A method for improving accuracy of a uniform driving behavior analysis device based on secondary identification

Abstract

The application discloses a kind of unified driving behavior analysis equipment accuracy based on secondary identification method.For overcoming the problem that the prior art lacks secondary analysis to DSM uploaded video, leading to false alarm to driver's illegal operation, the application includes that behavior analysis equipment collects video data in the time interval before and after alarm time and uploads to secondary analysis platform;Secondary analysis platform processes video data to generate several slices, compares the score of corresponding slice with the preset score, determines and records abnormal early warning slice;Secondary analysis platform sets early warning calculation rule according to the type of abnormal early warning slice, and determines the correct alarm of corresponding video data combined with the number of abnormal early warning slices.The alarm video generated by DSM equipment is calculated by the unified algorithm of platform secondary identification, and then the alarm is presented to the transportation management department;The application confirms through platform secondary identification, ensures unified algorithm and improves the accuracy of dangerous driving behavior identification.

Description

A method for improving the accuracy of unified driving behavior analysis devices based on secondary recognition Technical Field

[0001] This invention relates to the field of transportation, and in particular to a method for improving the accuracy of a unified driving behavior analysis device based on secondary recognition. Background Technology

[0002] To reduce the probability of traffic accidents caused by driver violations such as fatigue, smoking, and using mobile phones while driving, the transportation industry typically installs driver behavior analysis (DSM) devices in vehicles. These devices are capable of identifying driver violations such as fatigue, smoking, and using mobile phones while driving, and are generally installed directly in front of the driver's seat. When abnormal situations such as driver fatigue, smoking, or using mobile phones are detected, an alert is generated at the device's end, and a video clip of the alarm is transmitted back to the control center. However, due to differences in algorithms among DSM devices from different manufacturers, false alarms are frequent; furthermore, the lack of relevant verification devices makes them difficult for transportation management departments to use.

[0003] For example, a Chinese patent document, "A Method, System, Device, and Storage Medium for Analyzing Driver Behavior," with publication number CN2022113259875, includes collecting head and body movement image data of the driver while driving, identifying various driving behaviors, scoring these behaviors, calculating a comprehensive score based on weighted scores, and classifying the driver's behavior based on the comprehensive score. By collecting head and body movement image data of the driver while driving and identifying various driving behaviors based on this data, the accuracy of driver behavior recognition can be effectively improved. Furthermore, by scoring various driving behaviors and calculating a comprehensive score based on weighted scores, the driver's behavior can be classified. However, this solution lacks a secondary recognition device, and errors are easily uploaded directly when recognition fails. Summary of the Invention

[0004] The purpose of this invention is to address the problem of false alarms for driver violations caused by the lack of secondary analysis of DSM uploaded videos in existing technologies; it provides a method for improving the accuracy of unified driving behavior analysis devices based on secondary recognition. The alarm video generated by the DSM device is calculated by a unified algorithm for secondary recognition on the platform before the alarm is presented to the transportation management department; this invention ensures the unified algorithm and improves the accuracy of identifying dangerous driving behaviors through platform secondary recognition confirmation.

[0005] To achieve the purpose of this invention, the following technical solution is proposed to solve the problems of the prior art:

[0006] This invention includes: a secondary analysis platform receiving warning-related video data, processing the video data to generate several segments, comparing the scores of the corresponding segments with preset scores at a predetermined ratio, identifying and recording abnormal warning segments; the secondary analysis platform setting warning calculation rules based on the type of abnormal warning segments, and determining the correct alarm for the corresponding video data based on the number of abnormal warning segments. By performing secondary processing and verification on videos uploaded after an alarm from a behavior analysis device, correct alarm information can be obtained, thereby increasing the accuracy of abnormal warning review.

[0007] Preferably, the process of generating the warning-related video data includes: the behavior analysis device identifying the outline data of the detected object's illegal actions, issuing an alarm at the corresponding time, and collecting video data within a preset time interval before and after the alarm time to the secondary analysis platform. After identifying the outline data of the illegal actions, an alarm operation is performed. This alarm is transmitted to the secondary analysis platform to facilitate subsequent verification and ensure the smooth operation of the alarm.

[0008] Preferably, the behavior analysis device performs contour recognition on the detected object. The portion of the object's contour exceeding a preset threshold is designated as over-threshold data. Weights are calculated in a spatial coordinate system; if the weight exceeds a predetermined preset weight, it is identified as contour data of a violation. Standard driving contour data is stored in a pre-defined cloud database as a threshold, and various predicted violation contour data values ​​are used as references. Contour data exceeding the threshold range are then identified after calculation and compared with the reference data to confirm the violation. This method effectively ensures the accuracy of contour recognition.

[0009] Preferably, the behavior analysis device sets a discrimination type for the violation action contour data based on the difference between the violation action contour data and the preset conventional action contour data in the spatial coordinate system, and records an alarm for each violation action contour data corresponding to the discrimination type. By calculating the contour in each vector direction in the coordinate system and combining it with reference data within the system, the type of violation action can be effectively determined, improving the accuracy of the determination.

[0010] Preferably, the secondary analysis platform identifies the contour data of the detected object's illegal actions within the slice, scores the portion of the illegal action contour data that exceeds a preset threshold, and sets a preset score range of n% to m%. Slices with scores exceeding m% are recorded as danger warning slices under the abnormal warning slice category; slices with scores between n% and m% are recorded as risk warning slices under the abnormal warning slice type. For risk warning slices, a behavioral warning is issued; for danger warning slices, a specific alarm warning is issued. This effectively reminds drivers not to engage in illegal operations and also records any illegal operations that have already occurred.

[0011] Preferably, the secondary analysis platform sets the type labels for hazard warning slices as [A1, A2, ..., An] and the type labels for risk warning slices as [B1, B2, ..., Bn]. The same index corresponds to the same type of violation, with A representing a violation that has already occurred and B representing a violation that is predicted to occur. Setting type numbers helps to standardize the numbering of various types of violations and also facilitates the addition of new violation data, improving the overall database's adjustability and debugging flexibility.

[0012] Preferably, under the aforementioned early warning calculation rule, when the number of any abnormal early warning slice type exceeds the preset maximum value Qmax for a single type of abnormal slice, the alarm at the corresponding time is recorded as a correct alarm for that type of violation. This helps improve the accuracy of identifying single-type abnormal slices by ensuring correct alarm judgments for each type.

[0013] Preferably, under the aforementioned early warning calculation rule, the video data corresponding to a certain alarm contains several types of violations. If the number of abnormal early warning slices in the video data exceeds the preset maximum value Qnmax for multiple types of abnormal slices, then the alarm is recorded as a correct alarm for multiple types of violations. This method of judging correct alarms for multiple types helps improve the accuracy of identifying multiple types of abnormal slices.

[0014] The beneficial effects of this invention are:

[0015] 1. The present invention provides a method for improving the accuracy of a unified driving behavior analysis device based on secondary recognition. By using a unified algorithm standard, alarm information and video information uploaded using different DSMs are verified, and a unified standard driving behavior analysis result is output, which is beneficial for standardization and normalization.

[0016] 2. The present invention provides a method for improving the accuracy of a unified driving behavior analysis device based on secondary recognition. By leveraging the high computing power of the platform server, the analysis results of individual cameras are effectively filtered, thereby improving the accuracy of driving behavior prediction and analysis. At the same time, multiple discrimination results are set, which can play a role in predicting and alerting drivers to dangerous behaviors. Attached Figure Description

[0017] Figure 1 is a logic diagram of a method for improving the accuracy of a unified driving behavior analysis device based on secondary recognition according to the present invention.

[0018] Figure 2 is a schematic diagram of the method steps for improving the accuracy of a unified driving behavior analysis device based on secondary recognition according to the present invention. Detailed Implementation

[0019] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0020] Example:

[0021] This embodiment presents a method for improving the accuracy of a unified driving behavior analysis device based on secondary recognition, as shown in Figure 2. Its general steps include:

[0022] Step 1: The secondary analysis platform receives the warning-related video data, processes the video data to generate several slices, compares the scores of the corresponding slices with the preset scores according to a specified ratio, and determines and records the abnormal warning slices.

[0023] The process of generating warning-related video data includes: the behavior analysis device identifies the contour data of the detected object's illegal actions, issues an alarm at the corresponding time, and collects video data within a preset time interval before and after the alarm time to a secondary analysis platform. The behavior analysis device performs contour recognition on the detected object; the portion of the detected object's contour that exceeds a preset threshold forms over-threshold data. Weights are calculated in a spatial coordinate system, and if the weight is greater than a predetermined preset weight, it is identified as illegal action contour data.

[0024] Standard vehicle contour data is stored in a pre-defined cloud database as a threshold, and various predicted violation contour data values ​​are used as references. After calculation, data exceeding the threshold range is identified, and the violation action contour data is compared with the reference data to confirm the violation. This method effectively ensures the accuracy of contour recognition. The behavior analysis device sets a discrimination type for the violation action contour data based on the difference between the violation action contour data and the preset conventional action contour data in the spatial coordinate system. Violation action contour data corresponding to the discrimination type are recorded as an alarm. By calculating the contour in each vector direction in the coordinate system, combined with the reference data within the system, the type of violation action can be effectively determined, improving the accuracy of the discrimination.

[0025] A database adapted to this scheme should be provided to store historical reference contour data, standard contour data, and contour data measured on-site, facilitating subsequent comparison operations. Historical reference contour data should be used to establish detailed classifications of driver violations, including: driver fatigue, smoking, making or receiving phone calls, and other abnormal situations. Average contour data of other non-compliant behaviors should also be included as a detailed reference.

[0026] The coordinate system mentioned in the text should be a three-dimensional coordinate system adapted to this scheme. This coordinate system should include spatial variables in each direction. Based on the extension of the driver's profile data in each spatial variable direction measured on-site, an accurate determination of whether the profile information violates regulations can be made. Furthermore, the calculated results should be compared in detail with pre-set standard profile data. Any deviations should be weighted and the weighted calculation results compared with pre-referenced wheel library data to determine whether a violation has occurred, facilitating subsequent scoring operations.

[0027] The secondary analysis platform identifies the contour data of the detected object's violations within the slice. It scores the portion of the violation contour data that exceeds a preset threshold, setting a preset score range of n% to m%. Slices with scores exceeding m% are designated as danger warning slices under the anomaly warning slice category; slices with scores between n% and m% are designated as risk warning slices under the anomaly warning slice type. Danger warning slices are labeled with type labels [A1, A2, ..., An], and risk warning slices are labeled with type labels [B1, B2, ..., Bn]. The same index corresponds to the same violation; A represents a violation that has already occurred, and B represents a violation that is predicted to occur.

[0028] In practical applications, scoring is based on a weighted average, whereby the weighted average is used as the score. A minimum score percentage of n% is set; scores below this percentage are not considered violations (i.e., no obvious violations). Scores between n% and m% should be classified as risk warnings under anomaly alerts. Scores above m% constitute obvious violations and should be classified as danger warnings.

[0029] In addition, tools should be provided as prompts to provide drivers with alerts based on the video type corresponding to the risk warning type after the scoring is completed. These alerts should be delivered via voice prompts or display screens to clearly inform drivers that their current behavior is abnormal and urge them to stop immediately. Furthermore, drivers should be alerted to relevant videos that trigger hazard warnings, urging them to immediately cease dangerous behavior to ensure driving safety.

[0030] Step 2: The secondary analysis platform sets the warning calculation rules according to the type of abnormal warning slice, and determines the correct alarm for the corresponding video data based on the number of abnormal warning slices.

[0031] Under the early warning calculation rules, when the number of any abnormal warning slice type exceeds the preset maximum value Qmax for a single abnormal slice type, the alarm at the corresponding time is recorded as a correct alarm for that type of violation. The number of single-type abnormal slices is calculated using the platform's computing power. Once the specified number is exceeded, it is recorded as a correct alarm and sent to the review end, i.e., the operations and maintenance personnel's review, for processing as a minor violation. If the number is not exceeded, it should be handled in the same way as a danger warning, prompting the driver to stop the behavior and indicating that a dangerous behavior has occurred, while also sending it to the review end as a relatively minor violation.

[0032] Under the early warning calculation rules, if the video data corresponding to a certain alarm contains several types of violations, and the number of abnormal warning slices in the video data exceeds the preset maximum value Qnmax for multiple types of abnormal slices, then the alarm is recorded as a correct alarm for multiple types of violations. The number of abnormal warning slices for multiple types of violations is checked. If it exceeds the specified number, it is recorded as a correct alarm and sent to the review end for evaluation as an extremely serious violation. If it does not exceed the specified number, the driver should be prompted to stop the violation, and it should be sent to the review end as a relatively serious violation.

[0033] As shown in Figure 1, the complete operating logic of this scheme is as follows:

[0034] At the driver behavior analysis device, the device analyzes and monitors the driver's behavior. If the driver exhibits abnormal driving behavior, an alarm will be generated and the alarm video will be stored. The relevant video will then be sent to the secondary analysis platform. The secondary analysis platform first manually or automatically sets the scoring threshold and warning calculation rules. After receiving the video, it segments the alarm video by second and scores the probability of various abnormal behaviors in each segment.

[0035] The secondary analysis platform determines whether the probability of various abnormal behaviors is higher than the scoring threshold. If not, the image is not included in the early warning calculation rule. If it is, the image is recorded as included in the early warning calculation rule for that type of abnormal behavior. The platform then determines whether the total number of slices marked as having that type of abnormal behavior exceeds the early warning calculation rule. If it does, the slice is determined to contain that abnormal behavior, and the result that the image contains that type of abnormal behavior is output. If the determination is correct, an alarm is triggered. If the number of slices does not exceed the rule, the image is determined not to contain that type of abnormal behavior.

[0036] A practical example of this solution is as follows: The secondary analysis platform slices a ten-second alarm video and performs algorithmic judgment. Ten slices are obtained by slicing once per second. Each slice scores the probability of various violations, with a maximum score of 1. Assuming the thresholds for various alarms are set as follows: 4 instances of smoking within 10 seconds (0.88 or higher), 4 instances of making or receiving phone calls within 10 seconds (0.88 or higher), 1 instance of one or both hands leaving the vehicle within 10 seconds (0.65 or higher), and 4 instances of fatigued driving within 10 seconds (0.77 or higher), then the abnormal type of the load logic in the slice is determined to be "one or both hands leaving the vehicle." Other abnormal types are not met. The driver is warned for each abnormal type, and the abnormal behavior of "one or both hands leaving the vehicle" is reported to the maintenance personnel for record-keeping.

[0037] It should be understood that the embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

Claims

1. A method for improving the accuracy of a unified driving behavior analysis device based on secondary recognition, characterized in that, include: The behavior analysis equipment identifies illegal actions and triggers an alarm, collecting video data within a preset time interval before and after the alarm as warning-related video data. The secondary analysis platform receives the warning-related video data, processes the video data to generate several slices, compares the scores of the corresponding slices with the preset scores in a specified ratio, and identifies and records the abnormal warning slices. The secondary analysis platform sets early warning calculation rules based on the type of abnormal early warning slice, and determines the correct alarm for the corresponding video data based on the number of abnormal early warning slices.

2. The method for improving the accuracy of a unified driving behavior analysis device based on secondary recognition according to claim 1, characterized in that, The process of generating the warning-related video data includes: the behavior analysis device identifies the outline data of the illegal actions of the detected object, issues an alarm at the corresponding time, and collects video data of the preset time interval before and after the alarm time to the secondary analysis platform.

3. The method for improving the accuracy of a unified driving behavior analysis device based on secondary recognition according to claim 2, characterized in that, The behavior analysis device performs contour recognition on the detected object. The part of the detected object's contour that exceeds a preset threshold forms over-threshold data. Weights are calculated in the spatial coordinate system. If the weight is greater than the preset weight, it is identified as contour data of a violation.

4. The method for improving the accuracy of a unified driving behavior analysis device based on secondary recognition according to claim 2, characterized in that, The behavior analysis equipment sets a discrimination type for the violation action contour data based on the difference between the violation action contour data and the preset normal action contour data in the spatial coordinate system, and the violation action contour data corresponding to the discrimination type is recorded as an alarm.

5. A method for improving the accuracy of a unified driving behavior analysis device based on secondary recognition, as described in claim 1, 3, or 4, characterized in that... The secondary analysis platform identifies the contour data of the illegal actions of the detected object within the slice, scores the portion of the illegal action contour data that exceeds a preset threshold, sets a preset score range of n% to m%, and records the slices with scores exceeding m% as dangerous warning slices under the abnormal warning slice category. The slices with scores between n% and m% are denoted as risk warning slices under the anomaly warning slice type.

6. The method for improving the accuracy of a unified driving behavior analysis device based on secondary recognition according to claim 5, characterized in that, The secondary analysis platform sets the type labels of the hazard warning slices as [A1, A2, ..., An] and the type labels of the risk warning slices as [B1, B2, ..., Bn]. The same index corresponds to the same violation action. A represents the violation action that has already occurred, and B represents the violation action that is predicted to occur.

7. A method for improving the accuracy of a unified driving behavior analysis device based on secondary recognition, as described in claim 1 or 6, characterized in that... Under the aforementioned early warning calculation rule, when the number of any abnormal early warning slice type exceeds the preset maximum value Qmax of a single abnormal slice, the alarm at the corresponding time is recorded as a correct alarm for that type of violation.

8. The method for improving the accuracy of a unified driving behavior analysis device based on secondary recognition according to claim 7, characterized in that, Under the aforementioned early warning calculation rule, the video data corresponding to a certain alarm contains several types of violations. If the number of abnormal early warning slices in the video data exceeds the preset maximum value of multiple types of abnormal slices Qnmax, then the alarm is recorded as a correct alarm for multiple types of violations.

Images