TWI894931B - Electronic device and method for excluding misjudgment of driver violation behavior detection - Google Patents

Abstract

A method for detecting driving violations with a false positive elimination function, implemented by an electronic device, includes the following steps: (A) generating and storing a current image; (B) using a local object detection model to determine whether a driver has committed a driving violation and generating a local detection result. The method generates multiple local historical joint probabilities based on multiple historical images of the target captured within a predetermined time period and the current image. and a local current joint probability; (C) generating a local probability density function based on the local historical joint probabilities and the local current joint probability; (D) generating an error value based on the local probability density function and a reference truth probability density function; (E) determining whether the error value is less than a preset threshold; (F) if the error value is less than the preset threshold, determining that the local detection result is reliable; and (G) if the error value is greater than the preset threshold, determining that the local detection result is unreliable, and repeating step (A). A method for excluding misjudgment of driver violation behavior detection is using an electronic device. The method includes the following steps: (A) generating and storing a current image; (B) using a local object detection model to determine whether implemented a driver commits driving violation and generate a local detection result, and generating a plurality of local historical joint probabilities and a local current joint probability according to a plurality of target historical images captured during a predetermined time interval and the current image; (C) generating a local probability density function (PDF) according to the plurality of local historical joint probabilities and the local current joint probability; (D) generating an error value according to the local PDF and a ground truth PDF; (E) determining whether the error value is less than a preset threshold; (F) if yes, determining the local detection result is credible; and (G) if no, determining the local detection result is incredible and repeating step (A).

Description

A method and electronic device for detecting driving violations with a misjudgment elimination function

The present invention relates to a technology for detecting driving violations, and more particularly to a method and electronic device for detecting driving violations with a misjudgment elimination function.

Transportation companies use driver monitoring systems (DMS) to monitor and manage the driving behavior of their drivers, detecting violations such as drowsiness, distraction, or phone calls. DMS primarily utilizes in-car cameras to film the driver and analyze their behavior using image recognition technology. If any violations are detected, the incident is uploaded to the cloud.

In addition to detecting driver violations, vehicle driver monitoring systems also incorporate multi-tasking capabilities such as driving video recording, multi-camera image recognition, and event upload. Limited by hardware platform resource allocation, most vehicle driver monitoring systems utilize lightweight detection models or algorithms for image detection, analysis, and processing.

However, traditional methods for identifying violations often use a single-frame recognition threshold (e.g., a cell phone) to determine if a call is being made. Furthermore, lightweight detection models or algorithms often fail to account for unexpected environmental changes, leading to image recognition errors and misjudgments, resulting in false alarms or missed alarms. Furthermore, these errors can negatively impact the driver's record of violations. Frequent false alarms can also cause drivers to lose trust in the system.

Therefore, the purpose of the present invention is to provide a method for detecting driving violations with the function of eliminating misjudgments.

Therefore, the present invention has a driving violation detection method with a misjudgment elimination function, which is suitable for judging whether a driver has committed a driving violation while driving a vehicle. The method is implemented by an electronic device installed in the vehicle. The electronic device stores an image of the driver taken at a current time point and multiple historical images taken at past time points, and a method for judging whether the image contains multiple features. A local object detection model of an object and a ground truth probability density function pre-provided by a server are provided. The electronic device continuously photographs the driver to continuously generate and store images of the driver. The driving violation detection method includes step (A), step (B), step (C), step (D), step (E), step (F), and step (G).

In step (A), the electronic device captures the driver at a current time point to generate and store a current image.

In step (B), the electronic device utilizes the local object detection model to generate a plurality of local historical joint probabilities corresponding to the target historical images and a local current joint probability corresponding to the current image based on a plurality of target historical images captured within a predetermined time period in the historical images and the current image.

In the step (C), the electronic device generates a local probability density function related to the probability distribution of the driving violation occurring in the current image and the target historical images based on the local historical joint probabilities and the local current joint probability.

In the step (D), the electronic device generates an error value according to the local probability density function and the reference truth probability density function.

In the step (E), the electronic device determines whether the error value is less than a preset threshold.

In the step (F), when it is determined that the error value is less than the preset threshold, the electronic device generates a detection result indicating that the local detection result is reliable.

Another object of the present invention is to provide an electronic device for preventing misjudgment of driving violations.

Therefore, the present invention provides an electronic device for detecting driving violations with the ability to exclude false positives. The device is suitable for determining whether a driver has committed a driving violation while driving a vehicle. The device is installed in the vehicle and includes a storage unit, a camera unit, and a processing unit.

The storage unit stores a plurality of historical images of the driver taken at past time points, a local object detection model for determining whether the image contains a plurality of feature objects, and a ground truth probability density function.

The shooting unit is electrically connected to the shooting and storage unit for continuously shooting the driver so as to continuously generate and store images of the driver in the storage unit.

The processing unit is electrically connected to the photographing unit and the storage unit.

The shooting unit shoots the driver at a current time point to generate and store a current image in the storage unit. The processing unit uses the local object detection model to detect whether the driver has committed any illegal behavior, such as using a mobile phone to talk on the phone. Then, based on a plurality of target historical images shot within a predetermined time period in the historical images and the current image, a plurality of local historical joint probabilities corresponding to the target historical images and a local current joint probability corresponding to the current image are generated. The processing unit generates a local current joint probability corresponding to the current image. Based on the local historical joint probabilities and the local current joint probability, a local probability density function is generated, which is related to the probability distribution of the illegal behavior occurring in the current image and the target historical images. The processing unit generates an error value based on the local probability density function and the reference truth probability density function. The electronic device determines whether the error value is less than a preset threshold. When it is determined that the error value is less than the preset threshold, the electronic device generates a detection result indicating that the local detection result is reliable.

The present invention utilizes a local probability density function and a ground-truth probability density function to examine the error between the two and determine whether the error is less than a preset threshold. If the error is less than the threshold, the local object detection model's local detection results are reliable. If the error is not less than the threshold, the local object detection model's local detection results are unreliable. The electronic device then generates a new current image and performs new detection calculations. Image recognition and calculations can be re-performed in cases of doubt, reducing the likelihood of misjudgments and false alarms during driving violation detection.

Before the present invention is described in detail, it should be noted that similar elements are denoted by the same reference numerals in the following description.

Referring to FIG. 1 , the present invention is applicable to an embodiment of an electronic device 1 for implementing a method for detecting driving violations with a misjudgment elimination function. The device includes a storage unit 11, a communication unit 12, a camera unit 13, and a processing unit 14. These units can be electrically connected to transmit relevant information or data. The electronic device 1 is used to implement a method for detecting driving violations with a misjudgment elimination function. The misjudgment elimination function is applied to a local detection result generated by an object detection model, indicating whether a driver has engaged in a misjudgment while driving a vehicle, and determines whether the local detection result contains a misjudgment.

It is important to note that in this embodiment, the violation is, for example, talking on a handheld phone while driving, and the electronic device 1 is, for example, a vehicle driver monitoring system. In other embodiments, the violation may also be, for example, dozing off, without limitation. Therefore, for example, if the driver was not talking on a handheld phone while driving, but the local detection results indicate that the driver was talking on a handheld phone while driving, the above-mentioned false positive detection function will determine that the local detection results indicate a false positive.

The storage unit 11 stores a plurality of historical images of the driver taken at past time points, a local object detection model used to determine whether an image contains multiple characteristic objects, and a ground truth probability density function. The ground truth probability density function can be the factory setting of the electronic device 1 or received from the server during use after the electronic device 1 performs a calibration procedure and communicates with the server, but the present invention is not limited thereto.

The communication unit 12 is connected to a remote server 101 via a communication network 100. The remote server 101 stores a remote object detection model for determining whether an image contains multiple feature objects.

It is worth noting that in this embodiment, both the local object detection model and the remote object detection model are built using a deep learning approach. The local object detection model uses image recognition technology to generate a detection result indicating whether the driver has engaged in any illegal behavior while driving the vehicle. Examples of illegal behavior include, but are not limited to, dozing off or using a mobile phone. Generally speaking, due to considerations of vehicle interior space and the driver's line of sight, the electronic device 1 used to detect the driver's illegal behavior tends to be miniaturized. Consequently, the hardware contained within the electronic device 1 has limited performance, requiring only a digital signal processor (DSP) for computation. Consequently, the local object detection model can only perform a limited number of computational layers, and variables may only be integers, resulting in poor resolution of parameter values. In contrast, the remote server 101 is less hardware-restricted than the electronic device 1 capable of detecting driving violations and can use a graphics processing unit (GPU) for computation. Therefore, the remote object detection model can execute more computational layers, with more variables being floating-point numbers and parameter values having a higher resolution. Therefore, the remote server 101 utilizes higher-level comprehensive computational capabilities to employ a more complete and rigorously structured algorithm for image recognition in the remote object detection model, thereby obtaining detection results with a higher degree of reliability than the local object detection model.

It should be noted that in this embodiment, since the illegal behavior is to make a call with a handheld phone, the characteristic objects are, for example, a mobile phone, a hand, a face, and a mouth. In other embodiments, the characteristic objects may only be a mobile phone and a hand, but are not limited to this.

The shooting unit 13 is electrically connected to the storage unit 11 for continuously shooting the driver so as to continuously generate and store images of the driver in the storage unit 11, wherein the shooting area includes images of the driver's face and the area around the driver's seat, but is not limited thereto.

It should be noted that, in this embodiment, the photographing unit 13 photographs one image per second, but the present invention is not limited thereto.

The processing unit 14 is electrically connected to the storage unit 11, the communication unit 12, and the photographing unit 13.

Referring to Figures 1, 2, and 3, an embodiment of a method for detecting driving violations with a misjudgment elimination function is described. The following details the steps involved in this embodiment.

In step 201, the photographing unit 13 photographs the driver at a current time point to generate and store a current image in the storage unit 11.

In step 202, the processing unit 14 determines whether a calibration procedure is required. If it is determined that calibration is required, the process proceeds to step 203; if it is determined that calibration is not required, the process proceeds to step 207.

It is important to note that in this embodiment, the processing unit 14 determines whether calibration is necessary based on the current time and a preset period. When the processing unit 14 determines that the preset period has elapsed based on the current time, the processing unit 14 determines that calibration is necessary, but this is not a limitation. In this embodiment, the preset period is, for example, but not limited to, 10 minutes. If it starts at 09:00, calibration will be initiated at 09:10, 09:20, 09:30, and so on, every 10 minutes. In other embodiments, the electronic device 1 further includes a light sensor (not shown). When the light sensor detects changes in the light environment, such as when a vehicle enters or exits a tunnel, resulting in strong light or shadows cast around the driver's seat, the local object detection model may have an increased probability of erroneous image recognition results due to large changes in the external ambient light. Therefore, the amount of light in the environment can be confirmed by measuring the illumination around the driver's seat (unit: lux). When the detected illumination change is greater than a set value, the processing unit 14 determines that a calibration procedure needs to be executed. In addition, if the processing unit 14 detects that the facial features of the current image are different from those of the historical images, in order to avoid misjudgment when identifying driving violations of different drivers due to differences in the proportions of face shape, eyes, and mouth or facial features, the processing unit 14 determines that a correction procedure needs to be executed, but the present invention is not limited to this.

In step 203 , the processing unit 14 transmits a plurality of target historical images captured within a predetermined time period among the historical images and the current image to the remote server 101 via the communication unit 12 .

It is important to note that in this embodiment, the predetermined time period is from 10 seconds before the current time to 1 second before the current time. For example, if the current time is the 10th second, the predetermined time period is from 0 to 9 seconds, but this is not limited to this. Specifically, if the camera unit 13 is set to capture one image per second, the camera unit 13 will capture 9 images within the predetermined time period. The aforementioned number of images captured per second and the length of the predetermined time period are merely examples and are not intended to be limiting.

In step 204, the remote server 101 utilizes the remote object detection model and generates a remote detection result based on the received current image. The remote detection result indicates whether the driver has engaged in a driving violation. The remote detection result includes whether the driving violation has occurred and the probability values of each of the characteristic objects when the driving violation is present. The remote object detection model employs a comprehensive and rigorous algorithm to identify the current image. The remote server 101 generates a plurality of remote historical connection probabilities corresponding to the target historical images and a remote current connection probability corresponding to the current image according to the received target historical images and the current image.

The remote historical joint probability is generated by multiplying the probability values corresponding to the characteristic objects in each frame when the server 101 performs image recognition on the input target historical images based on the remote object detection model to generate the remote detection results. This generates multiple remote historical joint probabilities corresponding to each frame of the target historical image, representing the overall probability of a driving violation occurring in each frame of the target historical image. The remote current joint probability is generated in a similar manner to the remote historical joint probabilities, except that the image received and used is the current image, so this is not further described here.

In step 205, the remote server 101 generates a remote probability density function related to the probability distribution of the driving violation occurring in the target historical images and the current image within the predetermined time interval based on the remote historical joint probabilities and the remote current joint probability. Based on the remote probability density function, the remote server 101 calculates a plurality of real-time parameters related to the remote probability density function and transmits the real-time parameters to the communication unit 12 via the communication network 100. The communication unit 12 transmits the real-time parameters to the processing unit 14 via an electrical connection.

It is worth noting that in this embodiment, the real-time parameters include at least an event representative mean and an event representative variance. That is, when the remote server 101 uses the remote object detection model to perform image recognition on each frame of the target historical images and outputs the individual probabilities of the feature objects when they meet the driving violation behavior, the remote server 101 further generates a plurality of remote historical joint probabilities corresponding to each frame in the target historical images, and generates the remote current joint probability corresponding to the current image. Then, the remote probability density function is calculated using the remote historical joint probabilities and the remote current joint probability, and the remote probability density function is used to calculate the remote probability density function using a maximum likelihood estimation method. The real-time parameters that best fit the remote probability density function result within the time interval of the complete image data of the driving violation detection are estimated using MLE, namely, the event representative mean and the event representative variance, but not limited thereto.

In step 206, the processing unit 14 updates the baseline truth probability density function based on the real-time parameters.

In step 207, the processing unit 14 uses the local object detection model and the current image captured in step 201 to calculate a local detection result regarding whether the driver has committed a driving violation in the current image. The local detection result is also used to indicate whether the driver has committed a driving violation, and the local detection result includes whether the driving violation has occurred and the probability values of each of the characteristic objects when the driving violation is satisfied. The processing unit 14 also generates a plurality of local historical joint probabilities corresponding to the target historical images and a local current joint probability corresponding to the current image based on a plurality of target historical images captured within a predetermined time period in the historical images and the current image.

It is particularly noted that in step 204 and step 207 of this embodiment, the remote server 101 and the processing unit 14 both perform recognition based on images captured within the same predetermined time period. However, since the remote server 101 has better performance than the processing unit 14, in other implementations, the remote server 101 may perform image recognition in step 204 using images captured within a time period longer than the predetermined time period, while the processing unit 14 may also perform recognition using historical images of the target captured within the predetermined time period. In other words, the remote server 101 and the processing unit 14 may also use different predetermined time periods. The remote server 101 increases the accuracy of the overall judgment by calculating and analyzing image data over a longer period of time, while the processing unit 14 mainly uses the amount of data that can meet real-time computing needs. For example, the remote server 101 may perform recognition based on images captured over a 50-second period, while the processing unit 14 may only perform recognition based on images captured within a 10-second period. In this embodiment, the electronic device 1 inputs the current image into the local object detection model. The processing unit 14 calculates the probability values corresponding to the individual feature objects based on the local detection results of the local object detection model and multiplies them to generate a local current joint probability corresponding to the current image. Similarly, the processing unit 14 inputs the target historical images into the local object detection model. The processing unit 14 calculates the probability values corresponding to the feature objects to generate a plurality of local historical joint probabilities corresponding to each frame of the target historical images.

In step 208, the processing unit 14 generates a local probability density function related to the probability distribution of the driving violation occurring in the current image and the target historical images within the predetermined time interval according to the local historical joint probabilities and the local current joint probability.

Referring to Figure 4, an example of a local joint probability distribution is shown, consisting of the local current joint probability value corresponding to the current image captured between seconds 10 and 11, and the local historical joint probability values corresponding to the target historical images captured between seconds 0 and 10. The probability distribution of the local probability density function, the ground truth probability density function, and the remote probability density function is, for example, a normal probability density function (PDF).

In step 209, the processing unit 14 generates an error value between the local probability density function and the reference truth probability density function according to the local probability density function and the reference truth probability density function.

It should be noted that in this embodiment, the error value is the cross entropy of the baseline truth probability density function and the local probability density function. That is, the result of the cross entropy calculation is used to measure the degree of error between the electronic device 1 and the remote server 101 in detecting driving violations and their probability density distribution. If the results calculated by the two are close (high overlap), the cross entropy value will be lower (smaller difference), indicating that the possibility of uncertainty generated in the calculation data by the two is less, and therefore the credibility of the detection result of driving violations is higher, but the measurement method is not limited to this.

In step 210, the processing unit 14 determines whether the error value is less than a preset threshold. If the processing unit 14 determines that the error value is less than the preset threshold, the processing unit 14 generates a detection result indicating that the local detection result is reliable, and the process proceeds to step 211. If the processing unit 14 determines that the error value is not less than the preset threshold, the processing unit 14 generates a detection result indicating that the local detection result is unreliable, and step 201 is repeated.

In step 211, the processing unit 14 generates a local detection result regarding the driver's driving violation based on the local object detection model. When the error value is lower than the preset threshold, the processing unit 14 determines that the local detection result is reliable. If the local detection result indicates that a driving violation has occurred, the process proceeds to step 212; if the local detection result indicates that a driving violation has not occurred, the process proceeds to step 201.

In step 212, the processing unit 14 generates a warning message.

It is worth noting that the warning message can be a voice message. The processing unit 14 can issue the warning message through a speaker (not shown) to remind the driver to stop the current distracted driving behavior caused by using the mobile phone. The warning message can also be a text message. The processing unit 14 can transmit the warning message to the remote server 101 via the communication module to uniformly record and store multiple driving information data such as the time, location, and duration of the driver's driving violation as historical data on the driver's driving behavior in the fleet management function, but the present invention is not limited to this.

In summary, the processing unit 14 of the electronic device 1 generates the error value based on the local probability density function and the reference truth probability density function, and determines whether the error value is less than the preset threshold. When it is determined that the error value is less than the preset threshold, it indicates that the local detection result of the local object detection model is credible. When it is determined that the error value is not less than the preset threshold, it indicates that the local detection result of the local object detection model is not credible, and the processing unit 14 will regenerate a new The present invention utilizes the computing power of the remote server 101 to execute a calibration procedure, thereby obtaining real-time parameters that are more consistent with the current image data to update the baseline truth probability density function in the electronic device 1. By eliminating the error between the updated baseline truth probability density function and the local probability density function, some incorrect driving violation identification results are eliminated, thereby reducing the occurrence of misjudgments or false alarms, and thus can indeed achieve the purpose of the present invention.

However, the above description is merely an example of the present invention and should not be used to limit the scope of the present invention. All simple equivalent changes and modifications made within the scope of the patent application and the contents of the patent specification of the present invention are still within the scope of the present patent.

1: Electronic Device 11: Storage Unit 12: Communication Unit 13: Camera Unit 14: Processing Unit 100: Communication Network 101: Remote Server 201-212: Steps

Other features and functions of the present invention are clearly illustrated in the accompanying drawings, wherein: Figure 1 is a block diagram illustrating an embodiment of an electronic device for detecting driving violations with a misjudgment elimination function according to the present invention; Figure 2 is a flow chart illustrating steps 201-206 of an embodiment of a method for detecting driving violations with a misjudgment elimination function according to the present invention; Figure 3 is a flow chart illustrating steps 207-212 of the method for detecting driving violations with a misjudgment elimination function according to the present invention; and Figure 4 is a schematic diagram illustrating a local joint probability distribution consisting of a local current joint probability value corresponding to a current image, a plurality of local historical joint probability values corresponding to a plurality of target historical images, and a local probability density function.

Steps 201-206

Claims (10)

A method for detecting driving violations with a false positive elimination function is disclosed. The method is applied to detect a local detection result, detected by an object detection model, indicating whether a driver has engaged in a violation while driving a vehicle. The detection method is implemented by an electronic device installed in the vehicle. The electronic device stores a plurality of historical images of the driver taken at past time points, a local object detection model for determining whether an image contains a plurality of characteristic objects, and a ground truth probability density function. The electronic device continuously captures the driver to continuously generate and store images of the driver. The method for detecting driving violations comprises the following steps: (A) The electronic device captures the driver at a current time to generate and store a current image; (B) The electronic device utilizes the local object detection model to generate, based on a plurality of historical images of the target captured within a predetermined time period in the historical images and the current image, a plurality of local historical joint probabilities corresponding to the target historical images, and a local current joint probability corresponding to the current image; (C) The electronic device generates, based on the local historical joint probabilities and the local current joint probability, a local probability density function related to the probability distribution of the violation occurring in the current image and the target historical images; (D) The electronic device generates an error value based on the local probability density function and the ground truth probability density function; (E) The electronic device determines whether the error value is less than a preset threshold; and (F) When it is determined that the error value is less than the preset threshold, the electronic device generates a detection result indicating that the local detection result is reliable. The driving violation detection method with misjudgment elimination function described in claim 1 includes an electronic device connected to a remote server via a communication network. The remote server stores a remote object detection model for determining whether an image contains multiple feature objects. Between steps (A) and (B), the method further includes the following steps: (H) the electronic device determines whether calibration is required; (I) When calibration is determined to be necessary, the electronic device transmits the target historical images and the current image to the remote server via the communication network. The remote server generates a remote detection result using the remote object detection model and, based on the target historical images and the current image, generates a remote probability density function (PDF) corresponding to the probability distribution of the driving violation occurring in the current image and the target historical images. The remote server then calculates a plurality of real-time parameters consistent with the remote PDF and transmits these parameters back via the communication network. (J) The electronic device updates the ground truth PDF based on the real-time parameters. (K) When calibration is determined not to be necessary, step (B) is performed. As described in claim 2, the driving violation detection method with a misjudgment elimination function, wherein, in step (H), when the electronic device determines that the current time point has experienced a preset cycle, the electronic device determines that correction is required. As described in claim 1, the driving violation detection method with a misjudgment elimination function, wherein in step (B), the electronic device inputs the current image into the local object detection model and, based on the local detection results, multiplies the individual probability values of the characteristic objects in the current image when they meet the driving violation requirements to generate the local current joint probability corresponding to the current image. As described in claim 1, the driving violation detection method with a misjudgment elimination function, wherein in step (B), the electronic device inputs the target historical image into the local object detection model and, based on the image recognition results, outputs the individual probability values of the characteristic objects in each frame of the target historical image when they meet the driving violation requirements. Then, the individual independent corresponding probability values in each frame are multiplied to generate a plurality of local historical joint probabilities corresponding to each frame of the target historical image. An electronic device suitable for implementing a method for detecting driving violations with a false positive detection function is provided in a vehicle and comprises: A storage unit storing a plurality of historical images of a driver taken at past time points, a local object detection model for determining whether an image contains a plurality of characteristic objects, and a ground truth probability density function; A capturing unit electrically connected to the capturing and storing unit for continuously capturing the driver to continuously generate and store images of the driver in the storage unit; and A processing unit electrically connected to the capturing unit and the storage unit; The capturing unit captures the driver at a current time to generate and store a current image in the storage unit. The processing unit generates a local detection result using the local object detection model and, based on a plurality of historical images of the target captured within a predetermined time interval in the historical images and the current image, generates a plurality of local historical joint probabilities corresponding to the target historical images and a local current joint probability corresponding to the current image. The processing unit generates a corresponding local historical joint probability based on the local historical joint probabilities and the local current joint probability. The processing unit generates an error value based on a local probability density function (PDF) representing the probability distribution of the driving violation occurring in the current image and the target historical images and the ground truth PDF. The electronic device determines whether the error value is less than a preset threshold. When the error value is determined to be less than the preset threshold, the electronic device generates a detection result indicating that the local detection result is reliable. The electronic device as described in claim 6 further includes a communication unit electrically connected to the processing unit, the communication unit being connected to a remote server via a communication network, the remote server storing a remote object detection model for determining whether an image contains a plurality of feature objects, wherein the processing unit determines whether correction is required. When correction is determined to be required, the electronic device transmits the target historical images and the current image to the remote server via the communication network, and the remote server generates a correction using the remote object detection model. A remote detection result is generated, and based on the target historical images and the current image, a remote probability density function is generated, which is related to the probability distribution of the driving violation occurring in the current image and the target historical images. A plurality of real-time parameters that conform to the remote probability density function are calculated and then transmitted back via the communication network. The electronic device updates the reference truth probability density function based on the real-time parameters. When it is determined that no correction is required, the processing unit generates the local historical joint probabilities and the current joint probability. The electronic device as described in claim 7, wherein when the electronic device determines that the current time point has reached the preset period, the electronic device determines that calibration is required. An electronic device as described in claim 6, wherein the processing unit inputs the current image into the local object detection model, and the local object detection model outputs a plurality of individual probability values corresponding to the feature objects when they meet the driving violation requirements. The processing unit then multiplies the individual independent corresponding probability values to generate the local current joint probability corresponding to the current image. An electronic device as described in claim 7, wherein the processing unit inputs the target historical image into the local object detection model, and the local object detection model outputs a plurality of individual probability values corresponding to each frame of the target historical image for the characteristic objects when they meet the driving violation requirements. The processing unit then multiplies the individual independent corresponding probability values in each frame to generate a plurality of local historical joint probabilities corresponding to each frame of the target historical image.