TWI865049B - System for detecting occurrence period of cyclical event - Google Patents

Abstract

A system is configured for detecting an event occurrence period of cyclical event, and includes an image monitoring device and a period judgment device. The image monitoring device captures a plurality of monitoring images respective in a plurality of image capturing time. The period judgment device is configured to extract a plurality of feature vectors from the monitoring images to calculate a plurality of vector values to accordingly generate a vector value time domain signal; execute a short-time Fourier transform (STFT）with a window width time to transform the vector value time domain signal to a spectrogram; make the spectrogram be denoised to generate a representative frequency time domain signal; and calculate an event cycle achievement rate and a product term accumulation number after each window width time, so as to accordingly calculate event occurrence time and the event occurrence period.

Description

Periodic event detection system

The present invention relates to a detection system, and more particularly to a detection system for detecting the occurrence period of a periodic event.

As the industry develops towards automation and intelligence, unmanned and robotic production lines have gradually become the current mainstream development direction, and the production and quality monitoring work of the production lines has naturally been replaced by automated equipment.

Since the workpieces or products on the production line are arranged to perform periodic work according to the established work sequence, once the workpiece or product cannot complete the work that should be completed at one of the workstations of the production line according to the established work cycle, it will cause the work of the entire production line to be delayed. Therefore, appropriate devices are needed to monitor the work cycle of each workstation.

Looking at the existing work cycle detection technology, most of them rely on a large number of detection devices (such as light blocking end detection devices, touch detection devices, pressure sensing devices or chip detection devices, etc.) to monitor the position of the workpiece or the movement trajectory of the specific mechanism of the working equipment, and then combine the Internet of Things technology with the background big data computing platform to analyze and confirm whether the work to be completed at each workstation is completed within the allowed work cycle.

However, since the work content of each workstation is different, in order to fully realize the above technology, it is necessary to set up a variety of different detection devices to detect different parameters (such as temperature, pressure, number of touches, touch time, light-off signal interruption time, light-off signal delay time, light-off signal duration), etc. according to the actual work content of each workstation. In case of adjustment of the work content of the production line, the type and installation location of the selected detection device also need to be readjusted. In other words, the previous work cycle detection technology generally lacks universality and cannot be used in various workstations. Therefore, more resources must be spent to realize the detection of the work event cycle of multiple different workstations.

In view of the fact that in the prior art, there is a general lack of universality in event occurrence (work) cycle detection technology, which leads to the need to spend more resources to realize the problem of event occurrence cycle (work event occurrence cycle of workstation) detection of multiple different periodic events (work events of workstation); the main purpose of the present invention is to provide a new cycle detection technology, so that it can be widely used in detecting the event occurrence cycle of multiple different types of periodic events, making it particularly suitable for application in the work cycle detection of multiple different types of workstations.

One of the necessary technical means adopted by the present invention to solve the problems of the prior art is to provide a periodic event occurrence cycle detection system, which is used to detect an event occurrence cycle of a periodic event and includes a monitoring image capture device and a cycle judgment device.

The monitoring image capture device periodically captures a plurality of static monitoring image data corresponding to a monitoring field of view at a plurality of image capture times according to an image capture cycle, so as to concatenate them into a dynamic monitoring image data, and the periodic events occur within the monitoring field of view.

The cycle judgment device is communicatively connected to the monitoring image acquisition device, and is installed with a judgment program. After executing the judgment program, it includes a feature vector acquisition module, a signal generation module, a spectrum conversion analysis module, a noise filtering module, and an event occurrence cycle judgment module.

The feature vector acquisition module captures dynamic monitoring image data from the monitoring image acquisition device, and extracts multiple feature vectors of static monitoring image data corresponding to the image acquisition time from the dynamic monitoring image data through a mathematical operation model generated by deep learning. The signal generation module calculates the feature vector into multiple vector values arranged in the order of occurrence of the image acquisition time, so as to generate a vector value time domain signal according to the image acquisition time and the vector value.

The spectrum conversion analysis module is used to perform a short-time Fourier transformation on the vector value time domain signal with an analytical window width and time interval to generate a spectrum diagram, and the spectrum diagram includes a plurality of component frequencies corresponding to a plurality of analytical window widths and time intervals and a plurality of component frequency amplitudes of a plurality of component frequency signals. The noise filtering module filters out at least one noise signal corresponding to a part of the component frequencies, extracts the remaining component frequency signals corresponding to the remaining component frequencies as a plurality of effective component frequency signals, and sequentially extracts the component frequency corresponding to one of the effective component frequency signals with the largest amplitude as a representative frequency of each analysis window width time interval in each of the plurality of above-mentioned analysis window width time intervals, thereby generating a representative frequency time domain signal.

The event occurrence cycle judgment module calculates multiple event occurrence times based on the representative frequency time domain signal and the analysis window width interval to judge the event occurrence cycle.

Among the subsidiary technical means derived from the above necessary technical means, preferably, the spectrum conversion analysis module defines the largest one of the component frequency amplitudes as a maximum amplitude, and defines one of the component frequencies corresponding to the maximum amplitude as a main frequency; and the noise filtering module defines a noise critical amplitude according to the maximum amplitude, and regards the component frequency signals corresponding to the above-mentioned component frequencies whose component frequency amplitudes are smaller than the noise critical amplitude as the at least one noise signal and filters them out, wherein the noise critical amplitude can be 10% of the maximum amplitude. Preferably, the analysis window width and time interval can be equal to the image acquisition cycle, and the spectrum conversion analysis module uses the image acquisition cycle to perform short-time Fourier transformation.

The monitoring image capture device may be an IP Cam, and the periodic event may be a periodic work event of a production line, and the monitoring field of view is a monitoring field of view of a workstation of the production line. In addition, the periodic judgment device may further include a human-machine interface module for a user to set a focus area in the monitoring field of view, and the periodic event occurs periodically in the focus area. Preferably, the human-machine interface module includes a display and an operation interface. The display displays the dynamic monitoring image data of the monitoring field of view. The operation interface is for the user to set the focus area.

The feature vector acquisition module may further include a focus region feature vector acquisition unit, and the focus region feature vector acquisition unit uses a mathematical operation model to acquire a feature vector of static monitoring image data corresponding to the image acquisition time from the portion of the dynamic monitoring image data corresponding to the focus region.

The event occurrence cycle judgment module may further include a success rate calculation unit, a statistics unit, an event occurrence time calculation unit and an event occurrence cycle calculation unit. The success rate calculation unit accumulates the time interval frequency product terms in sequence to obtain an event cycle success rate. The statistics unit calculates a product term accumulation number every time the event success rate just reaches an integer, so as to calculate the corresponding multiple product term accumulation numbers when the event success rate just reaches multiple integers. The event occurrence time calculation unit multiplies the product term accumulation number by the analysis window width time interval to calculate multiple event occurrence times. The event occurrence cycle calculation unit calculates the event occurrence cycle based on at least one event occurrence time difference between any two adjacent event occurrence times.

In summary, the event occurrence cycle detection system of the periodic event provided by the present invention directly analyzes the event occurrence cycle of the periodic event from the static monitoring image data in the dynamic monitoring image data. Therefore, it can be directly used for the detection of the event occurrence cycle (such as the work event occurrence cycle of the work station) of multiple different periodic events (such as the periodic work events of the workstation). Therefore, it is no longer necessary to spend more resources to realize the effect of event occurrence cycle detection of multiple different periodic events.

Especially when it is applied to workstations, it is no longer necessary to spend more resources to easily detect the work event occurrence cycle of work events in multiple different workstations. Therefore, the execution cost of detecting multiple work event cycles in the production line can be greatly reduced.

100: Detection system

200: Workstation

210: Input conveyor belt

220:Working equipment

2201:Robot Arm

2202:Working platform

2203: Detection probe

230: Output conveyor belt

300: Circuit board to be tested

1: Monitoring image capture device

2: Cycle judgment device

21: Human-machine interface module

211: Display

212: Operation interface

JAP: Judgment Program

22: Feature vector extraction module

221: Focus on the regional feature vector acquisition unit

23:Signal generation module

231: Vector numerical operation unit

232:Signal generation unit

24: Spectrum conversion analysis module

25: Noise filtering module

251: Noise filtering unit

252: represents the frequency signal generation unit

26: Event occurrence cycle judgment module

261: Achievement rate calculation unit

262: Statistical unit

263: Event occurrence time calculation unit

264: Event occurrence cycle operation unit

The first figure is a schematic diagram showing a monitoring image capture device capturing dynamic monitoring image data of a monitoring field of view in a preferred embodiment of the present invention; the second figure A to the second figure E are static monitoring image data captured during a portion of the image capture time in an event occurrence cycle; the third figure is a functional block diagram of the event occurrence cycle detection system of a periodic event provided by the preferred embodiment of the present invention; the fourth figure The first figure shows the waveform of the vector value time domain signal generated in the preferred embodiment of the present invention; the fifth figure shows the spectrum diagram in the preferred embodiment of the present invention; the sixth figure shows the spectrum diagram after the noise signal is filtered in the preferred embodiment of the present invention; the seventh figure shows the waveform of the representative frequency time domain signal in the preferred embodiment of the present invention; and the eighth figure shows the corresponding curve diagram of the number of event occurrences and the time of event occurrence of the present invention.

Since the event occurrence cycle detection system of the periodic event provided by the present invention can be widely used to detect the event occurrence cycle of various different periodic events, and is particularly suitable for detecting the event occurrence cycle of periodic work events, its application level is quite broad, so it will not be described one by one here, and only one of the better embodiments of its application in the production line workstation is listed for specific explanation, and this embodiment is only used to conveniently and clearly assist in explaining the purpose and effect of the embodiment of the present invention.

Please refer to the first to third figures. The first figure is a schematic diagram showing a monitoring image capture device capturing dynamic monitoring image data of a monitoring field of view area in a preferred embodiment of the present invention; the second figure A to the second figure E are static monitoring image data captured at a portion of the image capture time in an event occurrence cycle. As shown in the first to third figures, a periodic event occurrence cycle detection system (hereinafter referred to as "detection system") 100 is used to detect an event occurrence cycle of a periodic event, and includes a monitoring image capture device 1 and a cycle judgment device 2. In this embodiment, the periodic event may be a periodic work event occurring at a workstation 200 in a production line.

The workstation 200 includes an input conveyor belt 210, a working device 220 and an output conveyor belt 230. The working device 220 includes a robot arm 2201, a working platform 2202 and a detection probe 2203. In this embodiment, the periodic work events occurring at the workstation 200 include the following steps in chronological order: using the input conveyor belt 210 to convey a circuit board 300 to be tested toward the working equipment 220; using the robot arm 2201 to pick up the circuit board 300 to be tested from the input conveyor belt 210; placing the circuit board 300 to be tested on the working platform 2202; extending the detection probe 2203 and contacting the circuit board 300 to be tested; retracting the detection probe 2203, and using the robot arm 2201 to move the circuit board 300 to be tested from the working platform 2202 to the output conveyor belt 230.

According to the working sequence of the workstation 200, periodic working events will occur periodically with an event occurrence cycle. The monitoring image capture device 1 can be an IP Cam installed in the workstation 200, and periodically captures a plurality of static monitoring image data corresponding to a monitoring field of view at a plurality of image capture times according to an image capture cycle, so as to concatenate them into a dynamic monitoring image data, and the periodic events occur in the monitoring field of view.

The so-called "image capture cycle" refers to the time interval between any two adjacent captures of static surveillance images, which is the reciprocal of the frame rate in the dynamic surveillance image data. In this embodiment, the frame rate is 30fps, which means that 30 static surveillance images are captured per second, which is equivalent to an image capture cycle of 1/30 second, about 0.033 seconds. In other words, the dynamic surveillance image data is formed by concatenating the static surveillance image data at a frame rate of 30fps.

The so-called "monitoring field of view" refers to the field of view of the monitoring image capture device 1 that can capture images. In this embodiment, since the periodic event is a periodic work event occurring in the production line 200, the monitoring field of view is a monitoring field of view of a workstation of the production line 200, which covers part of the input conveyor belt 210, the working equipment 200 and part of the output conveyor belt 230, that is, the area corresponding to the diagram range of the second A figure to the second E figure. Based on the above, when the periodic work event proceeds to the above steps in chronological order, the static monitoring image data captured by the monitoring image capture device 1 includes the content shown in the second A figure to the second E figure.

The cycle judgment device 2 can be an industrial computer, a desktop computer, a laptop computer, a tablet computer or a smart phone, which is connected to the monitoring image acquisition device 1 by wired communication or wireless communication means, and includes a human-machine interface module 21, and is installed with a judgment program JAP. The human-machine interface module 21 includes a display 211 and an operation interface 212. The display 211 can display dynamic monitoring image data of the monitoring field of view. The operation interface 212 can be a touch panel, a button or a mouse, etc., for a user (operator of the cycle judgment device 2) to set a region of interest ROI in the monitoring field of view.

Generally speaking, the region of interest ROI can be selected from the region where the periodic changes in the dynamic monitoring image data are more significant when a periodic event occurs, or the region that can simultaneously cover the locations of multiple different periodic events, that is, the region that best reflects the periodic changes of periodic events. In this embodiment, the region that best reflects the periodic changes of periodic events is at the work platform 2202, so the region surrounded by the peripheral contour close to the work platform 2202 can be selected as the region of interest ROI.

After executing the judgment program JAP, the cycle judgment device 2 may further include a feature vector acquisition module 22, a signal generation module 23, a spectrum conversion analysis module 24, a noise filtering module 25 and an event occurrence cycle judgment module 26.

The feature vector extraction module 22 includes a focus region feature vector extraction unit 221. The feature vector extraction module 22 can directly extract multiple feature vectors of static monitoring image data corresponding to the image capture time from the dynamic monitoring image data through a mathematical operation model generated by deep learning. More preferably, the focus region feature vector extraction unit 221 of the feature vector extraction module 22 can also be used to extract multiple feature vectors of static monitoring image data corresponding to the image capture time from the part of the dynamic monitoring image data corresponding to the focus region ROI through another mathematical operation model generated by deep learning.

Regarding the mathematical operation model, the relevant parameters such as chromaticity, saturation, grayscale, sharpness, saturation, etc. of each pixel can be used as the input data for initial learning, and the initial mathematical operation model can be established by using neural networks, regression operations or other existing learning algorithms. Then, multiple versions of the initial mathematical operation model are modified through multiple deep learning updates. The so-called feature vector is a number of representative feature variables extracted by the mathematical operation model for each (each frame) static monitoring image data, and each feature variable is regarded as a multidimensional vector composed of a component.

For example, the mathematical operation model can extract four representative feature variables A ₁ , B ₁ , C ₁ and D ₁ for the first frame of static monitoring image data, and the feature vector corresponding to the first frame of static monitoring image data is ( A ₁ , B ₁ , C ₁ , D ₁ ). Similarly, the mathematical operation model can extract four corresponding representative feature variables A ₂ , B ₂ , C ₂ and D ₂ for the second frame of static monitoring image data, and the feature vector corresponding to the second frame of static monitoring image data is ( A ₂ , B ₂ , C ₂ , D ₂ ), and so on. In fact, the number of representative eigenvalues may be several, dozens, or even hundreds, depending on the calculation results of the mathematical operation model obtained after multiple learning and modification. The number of eigenvalues calculated by each different mathematical operation model may also be different.

It must be emphasized that since periodic events occur periodically, multiple static monitoring image data may undergo periodic changes. Therefore, no matter which mathematical operation model is used to extract feature vectors from multiple static monitoring image data, these feature vectors will inevitably show periodic changes. Therefore, no matter which mathematical operation model is used to generate deep learning, the present invention can be implemented.

In this embodiment, the image capture cycle is 1/30 second, that is, the image capture time of the first (first frame) static monitoring image data _is 1/30 second, and its corresponding feature vector is ( A1 , B1 , C1 _{, D1} ) ; _the image capture time of the second (second frame) static monitoring image data _{is 2/30} second _{, and} its corresponding feature vector is ( A2 , B2 , C2 , D2 ) _, and so on.

，特徵向量為(A ₂,B ₂,C ₂,D ₂)之向量數值V ₂即等於

。 The signal generation module 23 includes a vector numerical operation unit 231 and a signal generation unit 232. The vector numerical operation unit 231 calculates the feature _vector into a plurality of vector values arranged in the order of occurrence of the image capture time. The vector value can be calculated by square root addition. For example _, the vector value _V1 of the feature vector ( A1 , B1 , C1 , D1 ) _is equal _to

, the vector value V ₂ of the eigenvector ( A ₂ , B ₂ , C ₂ , D ₂ ) is equal to

.

The digital signal generating unit 232 can generate a vector digital time domain signal according to the image capture time and the vector digital value. For example, the first frame time (1/30 second) corresponds to the vector digital value V ₁ , the second frame time (2/30 second) corresponds to the vector digital value V ₂ , and so on. Through the correspondence between the vector digital value and time, a vector digital time domain signal can be generated, and its waveform is shown in the fourth figure.

The spectrum conversion analysis module 24 is used to perform a short-time Fourier Transform (STFT) on the vector valued time domain signal with an analytical window width to generate a spectrum diagram (as shown in the fifth figure), and the spectrum diagram includes a plurality of component frequencies corresponding to a plurality of analytical window widths and a plurality of component frequency amplitudes of a plurality of component frequency signals. The short-time Fourier Transform (STFT) is to cut the vector valued time domain signal into a plurality of local signals with a time interval of the analytical window width, and then perform Fourier transformation on the plurality of local signals. Therefore, the resolution window width interval should be an integer multiple of the image acquisition cycle (1/30 second), preferably 1 times, that is, the resolution window width interval is exactly equal to the image acquisition cycle.

The spectrum conversion analysis module 24 can define the largest one of the component frequency amplitudes as a maximum amplitude for each analysis window width time interval, and define one of the component frequencies corresponding to the maximum amplitude as a main frequency. In this embodiment, the first analysis window width time interval is 0 to 1/30 seconds, and the second analysis window width time interval is 1/30 to 2/30 seconds. In this embodiment, the main frequency is approximately 1/70 Hz (approximately equal to 0.014 Hz).

The noise filtering module 25 includes a noise filtering unit 251 and a representative frequency signal generating unit 252. The noise filtering unit 251 can define a noise critical amplitude according to the maximum amplitude, and regard the above-mentioned component frequency signals corresponding to the component frequency amplitudes less than the noise critical amplitude as the at least one noise signal and filter them. Preferably, the noise critical amplitude can be 10% of the maximum amplitude. After the noise is filtered by the noise filtering unit 251, the spectrum diagram shown in the fifth figure will be transformed into the spectrum diagram shown in the sixth figure.

The representative frequency signal generating unit 252 can extract the remaining component frequency signals corresponding to the remaining component frequencies as a plurality of effective component frequency signals, and sequentially extract the component frequency corresponding to one of the effective component frequency signals with the largest amplitude in each analysis window width time interval as a representative frequency in each analysis window width time interval, thereby generating a representative frequency time domain signal. That is, the frequency with the largest amplitude in the first analysis window width (0~1/30 seconds) is captured as the representative frequency of the first analysis window width (0~1/30 seconds), and the frequency with the largest amplitude in the second analysis window width (1/30~2/30 seconds) is captured as the representative frequency of the second analysis window width, and so on, and the representative frequency time domain signal waveform shown in Figure 7 can be obtained. Because each analysis window width corresponds to a representative frequency, the representative frequency time domain signal waveform will be a square wave with amplitude changes. In this embodiment, the main frequency is about 1/70Hz (about 0.014Hz). As can be seen from Figure 7, the representative frequency corresponding to most of the analysis window width intervals is 1/70Hz (about 0.014Hz).

The event occurrence cycle judgment module 26 may include an achievement rate calculation unit 261, a statistics unit 262, an event occurrence time calculation unit 263 and an event occurrence cycle calculation unit 264.

The achievement rate calculation unit 261 multiplies each of the analysis window width time intervals with the corresponding representative frequency to calculate a time interval frequency product term, and sequentially accumulates the time interval frequency product terms to obtain an event cycle achievement rate. Specifically, the event cycle achievement rate (ECAR) can be calculated by the following formula:

Wherein, i represents the i-th analysis window width time interval; fi represents the representative frequency corresponding to the i-th analysis window width time interval, _and the unit is Hz; T _W represents the analysis window width time interval, which is 1/30 second in this embodiment; and n represents the number of accumulated product items of the time-frequency product items.

The event cycle achievement rate ECAR equals 1, which means that one periodic event is completed; the event cycle achievement rate ECAR equals 2, which means that two periodic events are completed, and so on.

The statistical unit 262 counts a product term accumulation number every time the event achievement rate just reaches an integer, so that when the event achievement rate just reaches a plurality of the above integers, the corresponding plurality of the above product term accumulation numbers are respectively counted. In this embodiment, when the product term accumulation number n is approximately equal to 2100, 4201, 6300..., the event cycle achievement rate ECAR just reaches 1, 2, 3... respectively. The so-called "just reaches" means "just equal to" or "if there is no exactly equal, the one that is greater than but closest is just reached".

For example: if the number of accumulated items n is 2100, the event cycle achievement rate ECAR=1, then the number of accumulated items n of the product items when the event cycle achievement rate ECAR just reaches 1 is 2100. Another example: if the number of accumulated items n of the product items is 2099, the event cycle achievement rate ECAR=0.9998; the number of accumulated items n of the product items is 2100, the event cycle achievement rate ECAR=1.0001; the number of accumulated items n of the product items is 2101, the event cycle achievement rate ECAR=1.0003, then the number of accumulated items n when the event cycle achievement rate ECAR just reaches 1 is 2100.

The event occurrence time calculation unit 263 multiplies the product term accumulation term by the analysis window width time interval respectively to calculate a plurality of event occurrence times. In this embodiment, when the event cycle achievement rate ECAR=1, the number of product terms accumulated n(2100) multiplied by the resolution window width interval (1/30) can be calculated to be 70 seconds for the first periodic event; when the event cycle achievement rate ECAR=2, the number of product terms accumulated n(4201) multiplied by the resolution window width interval (1/30) can be calculated to be 140.03 seconds for the second periodic event; when the event cycle achievement rate ECAR=3, the number of product terms accumulated n(6300) multiplied by the resolution window width interval (1/30) can be calculated to be 210 seconds for the third periodic event; and so on.

Since the event cycle achievement rate ECAR reflects the number of occurrences of periodic events, it can also be replaced by the number of occurrences of events, thus generating a corresponding curve diagram of the number of occurrences of events and the time of event occurrence as shown in Figure 8. Among them, each dot is a data point with an integer number of event occurrences. From left to right, the first dot indicates the time of occurrence of the first periodic event, the second dot indicates the time of occurrence of the second periodic event, and so on.

The event occurrence cycle calculation unit 264 can calculate an event occurrence time difference based on the time difference between the occurrence times of two adjacent events. Preferably, multiple event occurrence time differences can be averaged as the event occurrence cycle.

In this embodiment, the event occurrence time of the first periodic event is 70 seconds, the event occurrence time of the second periodic event is 140.03 seconds, and the event occurrence time of the third periodic event is 210 seconds. Therefore, the event occurrence time difference between the second periodic event and the first periodic event is 70.03 seconds, and the event occurrence time difference between the third periodic event and the second periodic event is 69.97 seconds. After arithmetic averaging the two, the event occurrence cycle can be obtained as 70 seconds. Alternatively, the remaining event occurrence time differences can be averaged with the above two event occurrence time differences to calculate a more accurate event occurrence cycle.

In practice, in addition to using the above method to determine a more accurate event occurrence cycle, the event occurrence cycle determination module 26 can also directly find the most representative frequencies corresponding to all analytical window width intervals in the representative frequency time domain signal waveform as the main frequency. In this embodiment, the most representative frequency is about 0.014Hz (equivalent to about 1/70Hz), indicating that the main frequency is about 1/70Hz, and then directly take the inverse of the main frequency to roughly estimate the event occurrence cycle to be about 70 seconds.

At the same time, in the eighth figure, the 30th dot mark falls at about 2100 seconds, indicating that 30 periodic events can be completed in about 2100 seconds, and it can also be verified that the event cycle is about 70 seconds. In addition, it can be seen from the seventh and eighth figures that the representative frequency of each analysis window width time interval is zero between 2100 seconds and 2600 seconds, and there are no dot marks, indicating that no periodic events occur during the period from about 2100 seconds to 2600 seconds.

Since the above-mentioned feature vector acquisition module 22, signal generation module 23, spectrum conversion analysis module 24, noise filtering module 25 and event occurrence cycle judgment module 26 are all generated by executing the judgment program JAP, the feature acquisition module 21, supervised learning module 22, judgment module 23 and verification alarm module 24 can essentially be the (partial) main program, sub-program of the judgment program JAP or the program page or function interface generated after executing the judgment program JAP. Anyone with general knowledge in the relevant technical field (especially the field of artificial intelligence algorithms) can use the above learning and judgment logic and appropriate programming language to write a judgment program JAP (including its main program or sub-program) with the above-mentioned eigenvector acquisition module 22, signal generation module 23, spectrum conversion analysis module 24, noise filtering module 25 and event occurrence cycle judgment module 26 functions, so as to realize the above-mentioned various technologies of the present invention.

In summary, since the event occurrence cycle detection system 100 provided by the present invention directly analyzes the event occurrence cycle of the periodic event from the static monitoring image data in the dynamic monitoring image data, it can be directly used for detecting the event occurrence cycle (such as the work event occurrence cycle of the workstation 200) of multiple different periodic events (such as the periodic working events of the workstation 200), so that it is no longer necessary to spend more resources to realize the effect of event occurrence cycle detection of multiple different periodic events.

Especially when it is applied to workstations, it is no longer necessary to spend more resources to easily detect the work event occurrence cycle of work events in multiple different workstations. Therefore, the execution cost of detecting multiple work event cycles in the production line can be greatly reduced.

The above detailed description of the preferred specific embodiments is intended to more clearly describe the features and spirit of the present invention, but is not intended to limit the scope of the present invention by the preferred specific embodiments disclosed above. On the contrary, the purpose is to cover various changes and arrangements with equivalents within the scope of the patent application for the present invention.

100: Detection system

1: Monitoring image capture device

2: Cycle judgment device

21: Human-machine interface module

211: Display

212: Operation interface

JAP: Judgment Program

22: Feature vector extraction module

221: Focus on the regional feature vector acquisition unit

23:Signal generation module

231: Vector numerical operation unit

232:Signal generation unit

24: Spectrum conversion analysis module

25: Noise filtering module

251: Noise filtering unit

252: represents the frequency signal generation unit

26: Event occurrence cycle judgment module

261: Achievement rate calculation unit

262: Statistical unit

263: Event occurrence time calculation unit

264: Event occurrence cycle operation unit

Claims (7)

A periodic event occurrence cycle detection system is used to detect an event occurrence cycle of a periodic event, and comprises: a monitoring image capture device, which periodically captures a plurality of static monitoring image data corresponding to a monitoring field of view area at a plurality of image capture times according to an image capture cycle, so as to concatenate them into a dynamic monitoring image data, and the periodic event occurs in the monitoring field of view area; and a cycle judgment device, which is communicatively connected to the monitoring image capture device, is installed with a judgment program, and executes the judgment program. The formula includes: a feature vector acquisition module, which acquires the dynamic monitoring image data from the monitoring image acquisition device, and acquires a plurality of feature vectors of the static monitoring image data corresponding to the image acquisition time from the dynamic monitoring image data by using a mathematical operation model generated by deep learning; a signal generation module, which calculates the feature vectors into a plurality of vector values arranged in the order of occurrence of the image acquisition time, so as to generate a vector value time domain signal according to the image acquisition time and the vector values; a spectrum conversion module An analysis module is used to perform a short-time Fourier transformation on the vector value time domain signal with an analytical window width and time interval to generate a spectrum diagram, and the spectrum diagram includes a plurality of component frequencies corresponding to a plurality of analytical window widths and time intervals and a plurality of component frequency amplitudes of a plurality of component frequency signals; a noise filtering module is used to filter out at least one noise signal corresponding to a portion of the component frequencies, extract the remaining component frequency signals corresponding to the remaining component frequencies as a plurality of effective component frequency signals, and sequentially extract the component frequency signals in the plurality of analytical windows mentioned above. For each of the wide time intervals, the component frequency corresponding to one of the effective component frequency signals with the largest amplitude is captured as a representative frequency of each of the analysis window width time intervals, thereby generating a representative frequency time domain signal; and an event occurrence cycle judgment module calculates a plurality of event occurrence times based on the representative frequency time domain signal and the analysis window width time interval for judging the event occurrence cycle, wherein the analysis window width time interval is equal to the image capture cycle, and the spectrum conversion analysis module performs the short-time Fourier transform with the image capture cycle. The event occurrence period detection system of a periodic event as described in claim 1, wherein the spectrum conversion analysis module defines the largest one of the component frequency amplitudes as a maximum amplitude, and defines one of the component frequencies corresponding to the maximum amplitude as a main frequency; and the noise filtering module defines a noise critical amplitude based on the maximum amplitude, and regards the component frequency signals corresponding to the above-mentioned component frequencies corresponding to the component frequency amplitudes smaller than the noise critical amplitude as the at least one noise signal and filters them out. The event occurrence periodicity detection system of a periodic event as described in claim 2, wherein the critical noise amplitude is 10% of the maximum amplitude, and the noise filtering module treats the above-mentioned part of the component frequency signals corresponding to the above-mentioned part of the component frequency amplitudes less than 10% of the maximum amplitude as the at least one noise signal and filters them out. The periodic event occurrence periodic detection system of claim 1, wherein the periodic judgment device further comprises a human-machine interface module for a user to set a focus area in the monitoring field of view, and the periodic event occurs periodically in the focus area. The event occurrence periodic detection system of a periodic event as described in claim 4, wherein the human-machine interface module comprises: a display for displaying the dynamic monitoring image data of the monitoring field of view; and an operation interface for the user to set the focus area. The event occurrence periodicity detection system of the periodic event as described in claim 1, wherein the feature vector acquisition module further comprises a focus region feature vector acquisition unit, and the focus region feature vector acquisition unit acquires the feature vectors of the static monitoring image data corresponding to the image acquisition time from the portion of the dynamic monitoring image data corresponding to the focus region by using the mathematical operation model. The event occurrence cycle detection system of the periodic event as described in claim 1, wherein the event occurrence cycle judgment module further comprises: an achievement rate calculation unit, which multiplies each of the analytical time intervals with the corresponding representative frequency to calculate a time interval frequency product term, and sequentially accumulates the time interval frequency product terms to obtain an event cycle achievement rate; a statistical unit, which calculates a product term accumulation term each time the event achievement rate is exactly an integer A number is used to calculate the corresponding cumulative number of product items when the event achievement rate just reaches a plurality of the above integers; an event occurrence time calculation unit is used to multiply the cumulative number of product items by the analysis window width time interval to calculate a plurality of event occurrence times; an event occurrence cycle calculation unit is used to calculate the event occurrence cycle based on at least one event occurrence time difference between any two adjacent event occurrence times.