WO2019081499A1 - METHOD AND EQUIPMENT OF DETONATION RECOGNITION BY SELECTIVE FILTERING - Google Patents

Abstract

One implementation of the invention relates to an item of equipment (10) for recognizing blasts caused by gunfire, supplied with power by an electric current source (1) and including a sound detector (2), which may be coupled to a sound amplifier (3), that is connected to a central data processing unit (4) incorporating a processor (40) in conjunction with rewritable non-volatile (51) and random-access (50) memories. In this item of equipment (10), the detected sounds are processed by the central unit (4) in conjunction with a continuous analysis module (6) including filters for discerning specific signatures of a spectral signal corresponding to a detected sound. The filters are activated in a continuous chain and, since the analysis is limited to the passing filters, the module (6) exhibits an overall latency time that is reduced to the latency time of the passing filters. Each filter transmits, as output, at least one specific signature conformity datum to the central unit (4), the filters of successive ranks connected in series having, at least for the first filters, run and latency times that increase progressively with their rank.

Description

 METHOD AND EQUIPMENT OF DETONATION RECOGNITION BY SELECTIVE FILTERING

DESCRIPTION

TECHNICAL AREA

 The invention relates to a method and equipment for detecting detonations caused by shots, in closed or open spaces, by selective filtering of these detonations.

 The detonation detonation, "detonation" being taken in the sense of propagation of a sound emitted by a gun-type source, falls within the field of security of sensitive places. Indeed, it aims to qualify the sounds of gunshot type other ambient noise, such as background noise from a playground or a gathering of people, such as a concert or event. This qualification relates to the type of sound emitted as well as their location, and this information is transmitted to centers or persons in order to possibly carry out the appropriate interventions.

 The sound signals are generally detected by several acoustic sensors. The nature of the sounds can then be evaluated by comparison between spectral signatures of the signals supplied by the sensors to the central unit and signatures of a database in connection with the central unit. A location of the sounds can also be performed by processing the signals picked up in the central unit according to the position of the sensors, for example by triangulation. STATE OF THE ART

This traditional detection is not satisfactory in the current context of varied background noise close to sounds emitted by firearms or explosive charges, such as firecrackers, fireworks, screaming, barking or applause. It is therefore difficult to discriminate, in a sufficiently selective, fast, reliable and reproducible way, the detonations of shots of simple or automatic weapons, or explosive charges, these ambient noise.

 [0005] Projects have been developed to improve the recognition of sounds. For example, US Pat. No. 5,612,729 discloses a method of producing a characteristic signature of an audio broadcast signal of broadcast channels. In this method, frequency bands are formed by conversion, each band representing portions of the audio signal transmitted in a predetermined frequency band of the broadcast channel. Then, a first group of these frequency bands is compared to a second group of these bands, the second group representing portions of the signal of which at least a portion was diffused before the portions of the signal represented by the corresponding portions of the first group. The signatures are then formed based on the comparison between the first and second groups. This solution can not be applied to detonation impulse detection which requires a treatment that is substantially more precise and more discriminating than a detection of radiofrequency signal waves that propagate in known frequency channels.

Furthermore, the patent document US 5,973,998 discloses a system for quickly locating shots or other explosive events. The system includes a battery of acoustic sensors mounted on the roofs or strategically useful positions. If a fourth signal confirms that an explosive event has occurred in a triangulated location from three other sensors, the system communicates this event to transmitting agents. These agents can view the pulse spectra of amplitude signals above a threshold. These pulses are converted to quadratic mean data (RMS), steep slope increase data, or fast Fourier Transform (FFT) frequency data to discriminate shots from other explosive events. This approach remains unreliable because the transmitted data are qualitatively compared to known sound profiles (barking, bird sounds, vehicle door closing) or threshold values. Moreover, these comparisons remain insufficient to characterize in a fine and sufficiently fast and reproducible way to provide relevant results that can be exploited to generate an intervention appropriate to the circumstances.

 In US patent document 2016133107, the closed-circuit fire detection system comprises sensors distributed in the enclosure under surveillance. The acoustic pulse of steep slope and high intensity can easily be distinguished from expected sound activities. An infrared detector can be added to capture the optical flash of a detonation in addition to the acoustic breath. Digital algorithms for qualifying an incident, based on the impulse characteristics of the sounds picked up, are advantageously interpreted by a specialist from amplitudes and waveforms, types of human sounds transmitted (cries or conversions, etc.). . It appears that this system is not efficient enough because it provides the possibility of intervention by a specialist to interpret the results.

STATEMENT OF THE INVENTION

 The invention aims precisely to reliably, quickly, reproducibly and selectively discriminate the detonations of gunshots ambient noise to allow an effective and proportionate intervention, to neutralize the space where such detonations are thus identified .

 To do this, the present invention provides that each sound event captured, and if necessary amplified, is analyzed by successive filtering recognition of a shot type detonation with zero or extremely low latency for the first time. filter or between filtering. And filtering that is not passing simultaneously and instantly results in a return to availability for the analysis of a next sound.

More specifically, the subject of the present invention is a method for recognizing detonations caused by shots, in which sounds of a given space, whether this space is open or closed, are detected and processed by filtering so that to discriminate said detonations other noises of different nature, for example ambient noise. This method comprises the following steps, after a preliminary calibration of the operating parameters of the filterings by a series of identical reference sounds: a detection is set up in said space and, a sound being detected, it is, if necessary after amplification, analyzed in continuous mode in the form of a signal formed of sound pulses by a series of successive evaluation filterings quantitative compliance with a reference detonation;

 the first filter performs a temporal analysis which presents a latency-free execution time, of shorter duration than the successive rankings that follow it and which perform conformity spectral analyzes establishing discrepancies between statistical data of signature extracted from sound pulses and corresponding signature data of the reference detonation with respect to standard conformance values;

 in this step, the statistical data extracted by the successive filterings correspond to processing execution times and latency times which increase, at least in the first filterings, filtering at the next higher rank filter while reducing the overall latency duration to the latency duration of the pass-through filtering, the overall filtering duration gradually increasing with the rank of the non-passing filter while remaining less than or equal to the execution time of all the filterings;

 the data of the pass-by filterings are recorded and the analysis is stopped as soon as a filtering is not running, the filtering being then immediately available for the sound detected, and if necessary amplified, which follows;

 - if all the filtering are passing, the analyzed sound is considered as resulting from a detonation produced by a shot.

 Under these conditions, the rejection of non-compliant sound detections is "accelerated" because the analysis time of the first filtering decreases from filtering to filtering that precedes in time, which also avoids an effect detection masking. The time saving thus achieved corresponds to the overall analysis latency reduced to the latency duration of pass-through filtering. The method makes it possible to reduce the electricity consumption compared to the state of the art which provides for the implementation of a prior sampling of the sounds before filtering.

According to preferred embodiments: the sound detection is omnidirectional and amplified, the amplification being composed of cascading stages, with an antireflection filtering stage which limits the highest frequencies of the detected sounds to a previously fixed threshold;

 the recognition of the sounds detected being carried out by digital filtering executed in sequence, the signals corresponding to the amplified sounds are converted into digital data for the analysis;

 the statistical data extracted by the successive filterings correspond to degrees of selectivity which increase, at least on the first filtering, filtering with rank filtering immediately higher proportionally to the corresponding latency duration, so that the most long running time are activated at the end of the analysis, a higher selectivity of a filtering in principle measuring its ability to provide a significantly higher number of positive responses over a given number of tests;

 the degree of selectivity increases on at least the first three filters;

 - the operating parameters of the filterings, determined during the calibration, may relate to the number of sound samples used, to the basic offsets of the signals coming from the sound detection so as to be analyzed on the same baseline, to the application of a sound level standardization factor of the data converted into digital blocks, the recording of the sound level of the data blocks for each relevant signal sample, the windowing and the normalization of the pulse signatures, the impulse thresholds taken into account for each sound detected, the sequence number of the block from which a signature is extracted, the extraction of the reference signatures, the number of activated filters and the tolerances allowed in the conformity analyzes;

an automatic selection of the amplification level is implemented by applying an optimized sensitivity level by a compromise between saturation and sufficient quality of the signal before conversion, by selecting the amplification stage from above level, compared to this floor of a sample of the signal transmitted to a mask between two thresholds of the voltage range allowed in the digital conversion, and to successively select the next lower stage if a comparison of the one-stage signal is outside said range and until that the signal is in the range of the lowest amplification stage;

 in a particularly advantageous embodiment, the analysis is performed by a succession of at least five filterings respectively comprising the following conformity analyzes: first signal level determination filtering corresponding to the detected sound with respect to a threshold predetermined; second filtering of determination of the power level of the signal by establishing a statistical average, this power corresponding to the sound volume of the detected sound; third envelope decay rate determination filtering of said signal in an interval around a predetermined reference value; fourth filtering for determining a gap level between envelope profiles of said signal and of a reference signal with respect to a predetermined threshold, and fifth filtering for determining a level of difference between the signatures of the spectra frequency of said signal and a reference signal with respect to a predetermined threshold;

 in the case where a filtering is on, the sound level extending over at least a first block of data is normalized and recorded;

 - In the second filtering, the sound volume is either fully recorded after acquisition of the entire sound signal, or partially in real time by a circular buffer memory reserved in sharp memory that frees the analysis when measuring a volume sound less than a corresponding reference volume threshold, which makes the filtering immediately off without waiting for the end of processing of the second filtering and reduces the latency of the analysis by the same amount;

in the third filtering, the slope of decay of the sound level is determined statistically from the sound levels of at least two data blocks and, if this slope is below a predetermined threshold, this filtering is conducting; in the fourth filtering, the difference between the envelope profiles is determined by summing statistics relating to the differences between the pulse levels for a given number of points of the spectral signal;

 in the fifth filtering, a comparison between the frequency distributions of the spectral signature of the signal and the reference signature is made to determine said difference, the frequencies of the filtered signal being determined from the data contained in at least a first block; ;

 the spectral signatures are determined in the fifth filter recursively by the product of previously normalized pairs of values of frequency and frequency intensity for each component of the frequency spectrum;

 in an additional downstream filtering of spectral decay analysis, the frequencies of the pulses of the spectral signal are divided into frequency bands and the filtering relates to a spectral density decay analysis in each of these bands, the end of the sound signal being determined by extrapolation of the progressive decrease in spectral density between the bands of higher and lower frequencies;

 - the signatures of a sound considered as a result of a gunshot are compared to isolated fire and / or firing dynamics signatures recorded on an experimental basis taking into account the time intervals between shots , the density of shots per sequence, and / or the diversity of the profiles of the signals, in order to deduce an identification of the source or sources of this detonation from these comparisons and the identification index of the detonation detected;

 the statistical data of the extracted signatures and conformance deviations are integrated in successive digital blocks of analysis of the filtering passing of higher rank;

the statistical data are transmitted via a bidirectional network to a terminal comparison application to isolated and / or dynamic reference shot signatures for identification purposes in a downstream analysis; the detection is put in standby mode in the absence of sound detection and is activated by a filter chosen between a digital detection of sound variation greater than a preset detection threshold in the first filtering and an analog detection by comparison with a predetermined threshold in a additional upstream filtering;

 in the standby mode, the sounds which precede a detection higher than the threshold of activation of the analysis are recorded and kept in a circular buffering memory reserved for random storage in order to be exploited in a downstream analysis;

 the spectral signal of the first data block is considered as background noise and is subtracted from the spectral signal of the last block in order to improve the quality of the filterings.

 The invention also relates to an equipment for detecting detonations caused by shots, powered by a source of electrical power and comprising at least one connected sound detector, where appropriate via an amplifier. of sound to which the sound detector is coupled, to a central data processing unit integrating a processor, in connection with at least one random access memory and a rewritable non-volatile memory, and at least one discriminating filter between said detonations and other noises. In this equipment, the sounds detected, and if necessary amplified, are processed by the central unit in connection with a continuous analysis module comprising signature discrimination filters specific for a spectral signal corresponding to a detected sound. The filters are activated in a continuous chain and, as the analysis is limited to pass filters, the module has a global latency duration reduced to the latency duration of the pass filters. Each filter transmits at least one signature compliance data item specific to the central unit, the series of successive row filters having, at least for the first three filters, latency times which progressively increase with their rank so that the overall latency time is reduced to the latency duration of pass-through filtering.

According to advantageous embodiments, the equipment comprises: at least one omnidirectional microphone, and an amplifier, connecting the sound detector to the central unit, composed of cascaded stages, with a prior non-folding stage constituted by a low-pass filter, each stage being coupled in output to the central processing unit via an analog / digital converter;

 the amplifier is connected to an input of an analog comparator, which is coupled to a potentiometer, so as to automatically adjust a detection threshold according to a sound environment; if this threshold is exceeded, the comparator activates the processor supply;

 the module comprises at least five following analysis filters: a first signal level determination filter corresponding to the detected sound with respect to a predetermined threshold; a second filter for statistical determination of the signal power level; a third envelope decay rate determination filter of said signal with respect to a predetermined threshold; a fourth filter for determining a gap level between envelope profiles of said signal and a reference signal with respect to a predetermined threshold, and a fifth filter for determining a difference level between signatures frequency spectra of said signal and a reference signal with respect to a predetermined distance;

 a random buffer reserved in RAM is coupled to the processor in the central processing unit, so as to perform a real-time processing of the sound detected in terms of volume so as to immediately release the sound volume discrimination filter in case sound level below a specified threshold;

 a differential pressure sensor is connected to the central processing unit in order to double the sound detection, this sensor being able to be chosen between a piezoelectric sensor, a capacitive pressure sensor and a microelectromechanical pressure sensor MEMS ;

a downstream processing remote processing unit is connected to the central processing unit via a local bidirectional network in order to transmit data frames containing statistical data established by the unit central from data produced by each filter and recorded in the memories;

 the downstream analysis relates to firing parameters selected from interval times between two shots, firing densities per firing sequence, sound profiles corresponding to different sources of firing, and / or a firing analysis; images recorded by a camera synchronously.

PRESENTATION OF FIGURES

 Other data, features and advantages of the present invention will appear on reading the following nonlimited description, with reference to the appended figures which represent, respectively:

 - Figure 1, 1e block diagram of an example of equipment according to the invention;

 FIG. 2, a graph of the execution times of five consecutive filters of an analysis module of the equipment of FIG. 1 as a function of the rank of the filter;

 FIGS. 3a to 3h, graphs of spectral signals analyzed by the filters of said module in order to extract conformance signatures from them and determine their deviations from reference signatures, and

 FIGS. 4a to 4e, graphs of a spectral signal of a recording of the sound produced by a source of shot highlighting respectively its spectral envelope (FIGS. 4a and 4b), its variances (FIG. 4c), its graph of the residuals after modeling by a linear regression (figure 4d) and its frequency spectrum after limitation and weighting (figure 4e).

DETAILED DESCRIPTION

 The block diagram of FIG. 1 relates to an example of equipment 10 for recognizing detonations caused by shots according to the invention. In this equipment 10, the main components are powered by an electric power source 1 and comprises an omnidirectional sound detector 2 coupled to a sound amplifier 3 which is itself connected to a digital data processing unit 4 integrating a processor 40.

In the central unit 4, memories 5 comprising random memories 50 and non-volatile rewritable 51 are connected to the processor 40, and an analysis module 6, consisting of a series of discriminating filters between detected sound detonations and other noises, in particular background noise. In the exemplary embodiment, the module 6 comprises five filters which are detailed below.

 In the equipment 10 the sounds detected by the detector 2 are first amplified by the amplifier 3 comprising three cascade stages 31 to 33, preferably preceded by an anti-aliasing low-pass filter frequencies above a given threshold. Moreover, this filter makes it possible to eliminate the artifacts in the spectra having undergone a Fourier transform. Each amplification stage 31 to 33 is connected at the output to the processor 40, advantageously via an analog multiplexer 7 coupled to an analog / digital converter 8 in this embodiment.

 In addition, each amplification stage 31 to 33 is advantageously connected to the input of an analog comparator 3C, which is coupled to a potentiometer 3P for automatic adjustment of the detection threshold of the detector 2, this potentiometer being digital in this example. Such an adjustment of the detection threshold is set according to the ambient sound environments at the time of sound recording. If this threshold is exceeded, the comparator 3C then activates the power supply of the processor 40, which makes it possible to generate energy savings.

 An automatic selection of the amplification level can be implemented by applying a sensitivity level optimized by a compromise between the saturation and a sufficient quality of the signal, by the following steps: selection of the stage higher level of amplification 33, comparison to this stage of a sample of the signal transmitted from the sound detection to a pulse template between two thresholds of the range of voltages allowed in the digital conversion; and successively selecting the next lower stage 32 if a comparison of the one-stage signal is outside said range, and until the signal is within the range of the lowest amplification stage, here the floor 31.

Preferably, a microelectromechanical differential pressure sensor (MEMS) 9 is connected to the central processing unit 4 in order to double the sound detection, parallel to the detector 2. This addition is particularly useful in a closed environment where a shot produces, in the volume thus limited, a sound component of very low frequency, for example less than 10 Hz, that the detector 2 does not perceive. not because of its performance generally limited in low frequency. The pressure detection obtained thus achieves an additional upstream filter, without increasing the detection latency of the analysis module 6. Alternatively, this pressure sensor may be a piezoelectric sensor or a capacitive pressure sensor.

 In addition, the central processing unit 4 is preferably connected to a remote downstream processing unit 1 1, via a secure local bidirectional network 12. This link allows to transmit the data frames containing statistical data. established by the central unit 4 from data produced by the analysis module 6 and recorded in the RAMs 50.

 Advantageously, a camera 13 records images synchronously with the central processing unit 4 in activated mode, and these images are transmitted to the remote processing unit 11 which integrates the analysis of these images into the image. downstream analysis of the statistical data transmitted by the central unit 4 in order to achieve an identification of the sources of detonations detected with a higher degree of certainty. The downstream analysis preferably relates to a set of parameters quantitatively related to the detected shots: interval times between two shots, density of shots per sequence to deduce the number of sources, diversity of sound profiles related to the variety of sources shooting, etc.

In operation, the sounds detected by the detector 2 and amplified by the amplifier 3 are processed by the central unit 4 in connection with the continuous analysis filter module 6. More precisely, and with reference to the graph time (as a function of time "t") of FIG. 2, each filter F1 to F5 of the analysis module 6 proceeds to the analysis, first temporally with the filter F1 and then spectrally with the other filters, of conformity of a specific signature of an upstream spectral signal "S" corresponding to a detected sound: the filter F1 determines the level of the signal "S" with respect to a detection threshold sO without latency for this filter; the filter F2 statistically determines the power level of the signal "S" corresponding to the sound volume of the detected sound; the filter F3 relates to the determination of the envelope decay rate of the signal "S"; the filter F4 determines the level of deviation, relative to a predetermined threshold, between the envelope profiles of said signal and of a reference signal, and the filter F5 relates to the determination of the level of difference between signatures frequency spectra of said "S" signal and a reference signal with respect to a predetermined threshold.

 At the end of the determination of the sound level of the filter F1, a conformity deviation datum is transmitted to the central processing unit 4 (arrow A1) and the analysis module 6 is then either in Waiting for the next spectral signal to be processed, the filter F1 being on, the analysis of the "S" spectral signal is immediately continued by transmitting the spectral signal "S" to the filter F2 (arrow P1). Similarly, at the end of the treatment of each of the filters F2 to F5, the conformity deviation datum is transmitted to the central unit 4 (arrows A2 to A5) and, in the case where the filter F2 to F4 is passing, the analysis is continued by instantaneous transmission of the spectral signal "S" to the filter F3 to F5 of rank immediately higher (arrows P2 to P4).

It is appropriate to define a latent time of a given filter as being the duration of execution of the equipment that precedes the activation of said filter. Under these conditions, the analysis module 6 has a global processing time reduced to the execution time of the activated filters, namely: T1 (for the filter F1), T1 + T2 (for the filters F1 and F2), T1 + T2 + T3 (for filters F1 to F3), T1 + ... + T4 (for filters F1 to F4) and, at most, equal to T1 + ... + T5 (for filters T1 to T5). In this example, a circular memory buffer reserved in RAM 50 (see FIG. 1) coupled to the processor 40 enables a real-time discrimination processing of the sound volume signature of the spectral signal "S", which makes it possible to release immediately the sound volume discrimination filter F2 without waiting for the end of the treatment, in case of sound level below a determined threshold. The use of such buffers can also shorten the processing times of other filters. In this exemplary embodiment, the filters of successive ranks F1 to F5 have execution times and thus latency that increase gradually with the rank of the filter. In other embodiments, it is possible for at least one filter of higher rank, for example the filter F4, to have a duration of execution and of latency less than a filter of lower rank, for example the filter F3. However, in all the embodiments, the first filters, for example the filters F1 to F3 with reference to FIG. 2, have execution and latency times that increase with their rank.

 FIGS. 3a to 3e illustrate graphs of the intensity "I" of spectral signals S1 to S5 in time "t", as analyzed by the five filters F1 to F5 of the analysis module 6, in function of the conformance signatures extracted from these signals, and their deviations from reference signatures. The signal scale is normalized to the higher intensity signal to compare sound events regardless of their absolute level and all spectral data is sampled and digitally recorded in the blocks.

 With reference to FIG. 3a, the detection level being the first conformity signature examined, a reference detection threshold sO is set beforehand, for example by automatic adjustment by the digital potentiometer 3P (see FIG. 1). In the example, the spectral signal S1 exceeds, for at least some pulses, the threshold s0, the filter F1 is declared as passing.

 More precisely, the absolute value of the spectral signal S1, minus the basic offset value due to the parameterization ("signal bias" in English terminology), is compared to a constant datum. The central processing unit 4 then records in its memories 50 (see FIG. 1) level data integrated in a vector composed of four blocks, B1 to B4. These blocks correspond to successive intervals of the same dimension in which the spectral signal S1 extends. The number and size of these blocks can be set by the operator.

With reference to FIG. 3b, the power level of the spectral signal S1, corresponding to the volume level of the sound initially detected, then constitutes the second conformity signature to be examined by the filter F2. This Power level is evaluated by determining a root mean square (or RMS) of sampled levels in the first block B1 which is transmitted to the central processing unit 4 (see Figure 1). If the RMS level is greater than a previously set reference threshold, the filter F2 is on and the sound power level of each data block B1 to B4 is recorded in the RAMs 50.

 The sound volume is either fully recorded after acquisition in the four blocks B1 to B4, or partially in real time, for example in the first block B1 as described above, by storage in a circular buffer reserved in memory long live 50 (see Figure 1). This memorization releases in particular the analysis when measuring a sound volume lower than a reference threshold, which makes the filter F2 immediately not running and reduces the execution time T1 + T2 by the same amount (see FIG. 2) of this analysis.

If the filter F2 is passing, the filter F3 determines statistically (see FIG. 3c) the slope Px corresponding to the decay rate of the spectral signal S1 during the sound detection which constitutes the third signature. The slope Px is evaluated statistically from RMS means at several points p, (p ₄ to ρβ) on at least one block of data, on the four blocks B1 to B4 in the exemplary embodiment. If the slope Px is in a given range, for example ± 10%, around a predetermined reference value P0, experimental or arbitrary, the filter F3 is passing. In another example, the spectral signal S2 of FIG. 3d has a statistically decreasing envelope with the time substantially shorter than that of the signal S1. The slope Py (FIG. 3e) is in the range validated around the reference value P0, and the filter F3 is passing.

With regard to the filter F4, it is a question of evaluating - with reference to FIG. 3f - the difference between the envelope profile of the spectral signal S1 with a predetermined reference envelope profile, experimentally or otherwise. arbitrarily. This difference, which constitutes the fourth conformity signature, is evaluated by the sumΣ of the squares of the deviations Δϊ (Δι to Δ3 on the graph of FIG. 3f) of points pj (p ₇ to pio) of the spectral signal S1 from a superposition (deviation Δ 0 0). If the sumΣ is less than a predetermined threshold, the filter F4 is passing.

 With reference to the graphs of FIGS. 3g and 3h, the filter F5 respectively analyzes the SF1 and SF2 spectrums of "F" frequencies of sound detections to determine the spectral signatures that constitute the fifth conformity signature. Thus, it appears that the first spectrum SF1 has a higher density of low frequencies, especially in the lower band BF1, while the spectrum SF2 contains a more significant density of high frequencies, in particular in the upper band BF2.

 Frequency spectra such as SF1 and SF2 are determined from the data contained in the first data block B1 (see FIG. 3a) by application of the Fast Fourier Transform (or FFT, acronym for "Fast Fourier Transform"). In English terminology). Then, the frequency values are used to evaluate the spectral signature by applying different modes of combination between the values of the spectral frequencies of the SF1 or SF2 spectra and their distribution.

 A combination mode consists in recursively determining the products of the pairs of values constituted by the coordinates of each component of the spectrum SF1 or SF2, respectively the frequency and the density of this frequency, and summing these products to obtain the spectral signature. If the difference between this spectral signature and the value of a reference signature is less than a predetermined value, the filter F5 is passing.

Other combination modes can be applied to determine a spectral signature from the coordinates (x, y) of points of the frequency spectra SF1 or SF2: sum of the products of the abscissae (x) by themselves, of these products (x) ² by the corresponding ordinate (y), abscissa (x) by the ordinates corresponding to the square (y) ² , and products of the abscissa squared (x) ² by the ordinate squared (y) ² . All these determination modes result in substantially the same deviation with the reference signature value determined by the same mode.

With reference to the graphs of FIGS. 4a to 4e, the spectral signal S3 illustrated (intensity of the signal as a function of time "t") relates to a detonation produced by a known firing source intended to serve as a reference in the analysis filters and during a preliminary calibration of the operating parameters: sampling frequency and number of samples used, determination of basic offsets subtracting spectral signals to normalize measurements by subtracting these offsets, determining a proportion factor between sound levels of data blocks to standardize data, applying a Hann-type windowing function, and normalizing pulses spectral signals, the sequence number of the block from which a signature is extracted, the extraction of reference signatures, the number of activated filters and the determination of tolerances in conformity analyzes, etc.

 To do this, it is necessary to isolate the portion of the signal S3 the most interesting which is located after the first third q1 of the signal S3 because in this first third, the signal S3 is generally partially saturated as it appears on the graph of Figure 4a. The duration of taking into account extends over an area in which the signal S3 has a significant intensity, that is to say on substantially the entire remaining portion q2 of the signal S3. For example, since the number of samples is a power of two because of the use of the FFT to determine the spectral signature in the F5 filter, the area of interest q2 is located at 1024 sampled records from the beginning of the signal. S3 and contains 4096 samples.

 The determination of the sound envelope Es of the signal S3 is illustrated by the graph of FIG. 4b. The envelope Es is obtained by iterative determination of the average RMS of sampled points of the signal S3 by limiting the detection threshold, the recording length by a compromise between a sufficient resolution of the profile and the number of sampled points. In the example, a record of 8192 samples was used and 64 records per data block. Such a determination is useful to serve as a reference at the level of the filter F4.

The envelope signature is reproducible and stable in the last two thirds of signal detection, as shown by the application of the following tools. By experimentally repeating a detonation produced by the same source under the same conditions, it appears that the graph of the variance of the signal V (t), as illustrated by Figure 4c, shows that the variance values "V" tend to rapidly reach zero, with a large peak at the beginning signal because of the saturation of this signal in this area. This result is confirmed by the application of a linear regression technique in the sense of the least, and to draw the graph of standardized residuals R (t). As illustrated in Figure 4d, this graph R (t) has a quantitative distribution little dispersed deviations.

 The spectrum SF3 of frequencies "F" of the reference shot that can be applied in the filter F5, appears in Figure 4e. This spectrum is limited by the isolated area according to Hann type windowing weighting.

 These studies also confirm the validity of the determination of the decay rate as a conformity signature in the filter F3, insofar as it appears that the rising edge (first third of the graphs of FIGS. 4a to 4d) remains substantially unreliable.

 The invention is not limited to the embodiments described and shown. Thus, the signature statistical data extracted by a filter of at least a given rank can be used by the rank filter that follows to establish the statistical data of the latter filtering, in particular for adjusting compliance values, and the data Signature statistics established by all the filters are processed globally by the central processing unit to provide an identification index of the detected detonation.

 Furthermore, the first analysis filter can integrate a first digital detection of sound variation greater than a preset detection threshold, which eliminates the analog comparator.

In addition, an downstream spectral decay analysis filter of the frequencies of the pulses of the spectral signal, in particular by Fast Fourier Transform or equivalent, divided into frequency bands may advantageously be added. This filter relates to the spectral density decay analysis in each of these bands, the end of the sound signal being determined by extrapolation of the progressive decrease in spectral density between the higher and lower frequency bands, in a Kalman type filter. In addition, the statistical signature data extracted by a filtering of at least one given rank can be advantageously used by the rank filtering which follows to establish the statistical data of the latter filtering, in particular to adjust values of compliance, and the signature statistical data established by all the filterings can also be processed globally, for example by the central processing unit, in order to provide an identification index of the detonation detected.

 The invention can also be applied to the recognition of burst fire from a repetitive recognition analysis shots encompassing the intervals between these shots.

 Alternatively to the self-adjustment of the detection threshold by the analog comparator coupled to the potentiometer, the first analysis filter can incorporate a first digital detection of sound variation greater than a preset detection threshold provided in this filter.

 In addition, the invention can be applied to the recognition of sound pulses in other contexts than that presented by the detonations of shots, in particular isolated or repeated pulses in contexts that can introduce noise. significant background, can be polluted by sounds of impulse characteristics close to those sounds to detect.

 In this context, the detection of sound detection related to a malfunction finds an application: valve malfunction detection in any type of fluid channeling circuit, relay in any type of electrical circuit, or expansion sensor in building or vehicle structures (automotive, aeronautical, railway, naval). Such detections make it possible to carry out predictive maintenance of this equipment.

It is also recommended to use the present invention in this context to detect, in a given environment, falls or displacements of objects in a given structure (in a building or a vehicle), or the presence and the counting pulses emitted by insects to quantify the amount of activity of these insects per unit of time. In this latter application, the ambient noise level can be advantageously recalculated automatically or periodically by the data processing unit. In addition, the cumulative evolution of the activity of the insects over time constitute curves presenting a linear portion of differentiated slope, making it possible to characterize the signature of one type of insect with respect to that of another and therefore, for example, to identify the type of insects present in a grain silo.

 Other applications are still within the scope of the present invention: detection of explosives, traffic accidents caused by a particular type of vehicle with two, three or four wheels, or by a fixed object (terminal , wall, etc.), noises in places of scientific experimentation, etc.

Claims

1. A method of recognizing detonations caused by shots, wherein sounds of a given space are detected and processed by filtering to discriminate said detonations of other noises of different nature, characterized in that it comprises the following steps , after a preliminary calibration of the operating parameters of the filterings by a series of identical reference sounds:  a detection is set up in said space and, a sound being detected, it is analyzed in continuous mode in the form of a signal formed of sound pulses (S1, S2) by a series of successive filterings of quantitative evaluation of conformity to a reference detonation;  the first filter performs a temporal analysis which has a latency-free execution time, of shorter duration than the successive rankings which follow it and which perform spectral conformity analyzes establishing discrepancies between the statistical signature data (Px , Py; Δί) extracted from the sound pulses and corresponding signature data of the reference detonation with respect to standard conformance values (sO, PO);  in this step, the statistical data extracted by the successive filterings correspond to processing execution times (T1 to T5) and latency times which increase, at least in the first filterings, a filtering with rank filtering immediately higher while reducing the overall latency duration to the latency duration of pass-through filtering, the overall filtering duration gradually increasing with the rank of the non-passing filter while remaining less than or equal to the execution time of all the filtering;  the data of the pass-by filterings are recorded (A1 to A5) and the analysis is stopped (P1 to P4) as soon as a filtering is not running, the filtering being then immediately available for the detected sound which follows;  - if all the filtering are passing, the analyzed sound is considered as resulting from a detonation produced by a shot. 2. Recognition method according to claim 1, wherein the sound detection is omnidirectional (2) and amplified, the amplification (3) being composed of cascading stages (31 to 33), with a preliminary low pass filter stage (30) which limits the highest frequencies of the detected sounds to a previously fixed threshold.  3. Recognition method according to any one of claims 1 or 2, wherein the recognition of detected sounds being performed by digital filtering executed in sequence, the signals corresponding to the amplified sounds are converted into digital data (8) for the analysis.  4. Recognition method according to any one of the preceding claims, wherein the statistical data extracted by successive filtering (F1 to F5) correspond to degrees of selectivity which increase, at least in the first filtering (F1 to F3), filtering at the next rank filter proportionally to the corresponding latency duration, so that the longest runtime filtering is activated at the end of the analysis.  5. Recognition method according to the preceding claim, wherein the degree of selectivity increases on at least the first three filters.  6. Recognition method according to any one of the preceding claims, wherein the operating parameters of the filterings, determined during calibration, relate to the number of sound samples used, the basic offsets of the signals from the detection. to be analyzed on a single baseline, the application of a noise level standardization factor of data converted to digital blocks (B1 to B4), windowing and signal normalization, and impulse thresholds. 7. Recognition method according to any one of the preceding claims, wherein an automatic selection of the amplification level is implemented by applying an optimized sensitivity level by a compromise between saturation and sufficient quality of the signal before conversion, by the selection of the higher level amplification stage (33), by comparison with this stage of a sample of the signal transmitted to a template between two thresholds of the range of voltages allowed in the digital conversion, and successively selecting the next lower stage (32) if a comparison of the one-stage signal is outside said range and until the signal is included in the range of the lowest amplification stage (31).  8. Recognition method according to any one of the preceding claims, wherein the analysis is performed by a succession of at least five filtering (F1 to F5) respectively comprising the following conformity analysis: first filtering level determination of signal (S1, S2) corresponding to the detected sound with respect to a predetermined threshold sO; second filtering of determination of the power level of the signal by establishing a statistical average, this power corresponding to the sound volume of the detected sound; third envelope decay rate determination filtering said signal (Px, Py) in a range around a predetermined reference value (PO); fourth filtering for determining a level of deviation between envelope profiles of said signal and of a reference signal (Es) with respect to a predetermined threshold, and fifth filtering for determining a level of difference between the signatures of the frequency spectra (SF1, SF2) of said signal and of a reference signal with respect to a predetermined threshold (SF3).  9. Recognition method according to the preceding claim, wherein, in the case where a filter is passing, the sound level extending over at least a first block (B1) of data is normalized and recorded.  10. Recognition method according to claim 8, wherein the sound volume analyzed by the second filtering is either fully recorded after acquisition of the whole of the sound signal, or partially in real time by a circular buffer memory reserved for random storage (5). ) which releases the analysis when measuring a sound volume below a corresponding reference volume threshold. 1 1. The recognition method according to claim 8, wherein the slope of decay of the sound level (Px, Py) is determined statistically in the third filtering from the sound levels of at least two data blocks (B1 to B4) and, if this slope is in the range around a predetermined reference value (PO), this filtering is on.  The recognition method according to claim 8, wherein the difference between the envelope profiles is determined in the fourth filtering by summing statistics relating to the differences (Δί) between the pulse levels for a number of points. (pi) given the spectral signal (S1).  13. Recognition method according to claim 8, wherein a comparison between the frequency distributions (BF1, BF2) of the spectral signature of the signal (SF1, SF2) and the reference signature (SF3) is established in the fifth filtering. for determining said difference, the frequencies of the filtered signal being determined from the data contained in at least a first block (B1 to B4).  14. Recognition method according to the preceding claim, wherein the spectral signatures are determined in the fifth filter recursively by the product pairs of previously standardized values of frequency and frequency intensity for each component of the frequency spectrum.  The recognition method according to any one of claims 8 to 14, wherein the frequencies of the pulses of the spectral signal are divided into frequency bands (BF1, BF2) in an additional downstream filtering of spectral decay analysis which relates to to a spectral density decay analysis in each of these bands, the end of the sound signal being determined by extrapolation of the progressive decrease in spectral density between the bands of higher (BF2) and lower frequencies (BF1). 16. A method of recognition according to any one of the preceding claims, wherein the signatures of a sound considered as resulting from a shot detonation are compared to signatures of isolated shots and / or shooting dynamics recorded on an experimental basis taking into account the time intervals between shots, the density of shots per sequence, and / or the diversity of signal profiles, in order to to deduce an identification of the source or sources of this detonation from these comparisons and the identification index of the detected detonation.  17. Recognition method according to any one of the preceding claims, in which the statistical signature data extracted by a filtering of at least one given rank are used by the rank filtering which follows to establish the statistical data of the latter filtering. , in particular for adjusting compliance values, and the statistical signature data established by all the filterings are processed globally to provide an identification index of the detonation detected.  18. Recognition method according to any one of the preceding claims, in which the statistical data of the extracted signatures and conformance deviations are integrated in successive digital blocks (B1 to B4) of analysis of the filtering passing of higher rank. .  19. A method of recognition according to any one of the preceding claims, wherein the statistical data are transmitted via a bidirectional network (12) to a terminal application (1 1) of comparison to isolated fire signatures and / or reference dynamics for identification purposes in a downstream analysis.  20. Recognition method according to any one of the preceding claims, wherein the detection is put in standby mode in the absence of sound detection and is activated by a filter chosen between a digital detection of sound variation greater than a predetermined detection threshold in the first filtering (F1) and an analog detection by comparison with a preset threshold in an additional upstream filtering (9).  21. Recognition method according to any one of the preceding claims, wherein in standby mode, the sounds which precede a detection higher than the activation threshold of the analysis are recorded and stored in a circular buffer memory reserved for random storage (5) in order to be exploited in a downstream analysis. 22. Recognition method according to any one of claims 18 to 21, wherein the spectral signal of the first block of data (B1) is considered background noise and is subtracted from the spectral signal of the last block in order to improve the quality of the filtering.  23. Equipment for detecting detonations caused by shots, supplied by an electric power source (1) and comprising at least one sound detector (2) connected to a central data processing unit (4) integrating a processor ( 40) in association with at least one random access memory (50) and a rewritable non-volatile memory (51), and at least one discriminating filter between said detonations and other noises, characterized in that in this equipment (10) ), the detected sounds are processed by the central unit (4) in connection with a continuous analysis module (6) comprising filters (F1 to F5) for discriminating specific signatures of a spectral signal (S1, S2 ) corresponding to a detected sound, in that the filters (F1 to F5) are activated in a continuous chain and, the analysis being limited to the passing filters, the module (6) has an overall latency duration reduced to the latency time (T1 to T5) filters p assants, and in that each filter (F1 to F5) outputs at least one specific signature compliance data to the central unit (4), the successive series-ranked filters having, at least for the first filters , execution times and latency (T1 to T5) which increase gradually with their rank.  24. Recognition equipment according to the preceding claim, wherein the equipment also comprises at least one omnidirectional microphone (2), and an amplifier (3), connecting the sound detector (2) to the central unit (4), composed plurality of stages (31 to 33) cascaded with a non-folding preconditioning stage constituted by a low-pass filter (30), each stage being coupled to the central processing unit (4) via a converter analog / digital (8). 25. Recognition equipment according to any one of claims 23 or 24, wherein the amplifier (3) is connected to an input of an analog comparator (3C), which is coupled to a potentiometer (3P), so to automatically adjust a detection threshold according to a sound environment, in case of exceeding this threshold the comparator (3C) activating the power supply (1) of the processor (40). 26. Recognition equipment according to any one of claims 23 or 24, wherein the first analysis filter (F1) of the module (6) incorporates a first digital sound variation detection greater than a preset detection threshold sO provided in this filter (F1) to activate the processor in standby (40).  27. Recognition equipment according to any one of claims 23 to 26, wherein the module (6) comprises at least five following analysis filters: a first filter (F1) signal level determination (S1, S2) corresponding to the detected sound with respect to a predetermined threshold sO; a second filter (F2) for statistical determination of the signal power level (S1, S2); a third filter (F3) for determining the envelope decay rate (Px, Py) of said signal (S1, S2) with respect to a predetermined threshold; a fourth filter (F4) for determining a gap level between envelope profiles of said signal (S1, S2) and a reference signal (S3) with respect to a predetermined threshold, and a fifth filter ( F5) for determining a difference level between signatures of frequency spectra (SF1, SF2) of said signal and of a reference signal (SF3) with respect to a predetermined distance.  28. Recognition equipment according to the preceding claim, wherein an additional downstream filter for spectral decay analysis of the frequencies of the pulses of the spectral signal (S1, S2) divided into frequency bands and the filter relates to a decay analysis of spectral density in each of these bands, the end of the sound signal being determined by extrapolation of the progressive decrease in spectral density between the higher and lower frequency bands.  29. Recognition equipment according to any one of claims 23 to 28, wherein a random buffer reserved memory buffer (50) is coupled to the processor (40) in the central processing unit (4), so that perform a real-time processing of the sound detected in terms of volume in order to immediately release the sound volume discrimination filter in case of sound level below a determined threshold. 30. Recognition equipment according to any one of claims 23 to 29, wherein a differential pressure sensor (9) is connected to the central processing unit (4) in order to double the sound detection, this sensor being able to be chosen between a piezoelectric sensor, a capacitive pressure sensor and a microelectromechanical pressure sensor MEMS.  31. Recognition equipment according to any one of claims 23 to 30, a downstream remote processing unit (1 1) is connected to the central processing unit (4) via a local bidirectional network (12) in order to transmit data frames containing statistical data established by the central unit from data produced by each filter (F1 to F5) and stored in the memories (5).  32. Recognition equipment according to any one of claims 23 to 31, wherein the downstream analysis (1 1) relates to shooting parameters selected from interval times between two shots, shooting densities per sequence of shots. shooting, sound profiles corresponding to varieties of firing sources, and / or an analysis of images recorded by a camera synchronously.