Method and device for tailgate forensics based on multi-source fusion

Abstract

This application relates to the field of vehicle-related technology and discloses a method and device for obtaining rear license plate evidence based on multi-source fusion. The method includes: in response to determining that a collision event has occurred, locking at least one responsible target from candidate targets continuously monitored by the vehicle's lidar before the collision; wherein the license plate of the responsible target is suspended at the rear; guiding an onboard camera to continuously track and image the location area indicated by the license plate location information provided by the lidar in real time, thereby obtaining a sequence of rear license plate images; identifying and verifying the sequence of rear license plate images; if valid license plate information of the responsible target is obtained before the responsible target escapes, then generating an evidence package uniquely corresponding to the responsible target based on the valid license plate information. Applying the technical solution of this application can solve the problem in the prior art where rear license plate evidence cannot be obtained in a very short time in a collision escape scenario.

Description

Technical Field

[0001] This application relates to the field of vehicle-related technology, specifically to a method and device for obtaining license plate information based on multi-source fusion. Background Technology

[0002] According to existing vehicle license plate standards, motorcycles are allowed to display only the rear license plate. In urban roads, delivery motorcycles, especially those used for food delivery and courier services, are a high-risk group for traffic accidents due to their frequent operation and high speeds. After a collision, motorcycles often skid or reverse at speeds of 12-20 m / s (approximately 43-72 km / h), with the escape process taking an extremely short time (usually ≤250 milliseconds).

[0003] However, existing evidence collection systems typically rely on the original vehicle's wide-angle rear camera with a frame rate of 30 frames per second (fps). Its low frame rate and poor dynamic capture capabilities result in a tail license plate recognition rate of less than 30%, making it difficult to complete tail license plate recognition before the escape and failing to meet the requirements for determining liability in traffic accidents. Summary of the Invention

[0004] In view of the above problems, this application provides a method and device for obtaining tail license plate evidence based on multi-source fusion, which is used to solve the problem in the prior art that tail license plate evidence cannot be obtained in a very short time in collision escape scenarios.

[0005] According to one aspect of the embodiments of this application, a method for obtaining evidence based on multi-source fusion of license plate numbers is provided, the method comprising:

[0006] In response to determining that a collision event has occurred, at least one responsible target is identified from candidate targets continuously monitored by the vehicle's lidar prior to the collision; wherein the license plate of the responsible target is displayed at the rear.

[0007] Based on the real-time tail license plate location information of the responsible target provided by the lidar, the vehicle-mounted camera is guided to continuously track and image the location area indicated by the tail license plate location information to obtain a tail license plate image sequence.

[0008] The image sequence of the last license plate is subjected to multi-frame recognition and verification. If the valid license plate information of the responsible target is obtained before the responsible target escapes, then an evidence package uniquely corresponding to the responsible target is generated based on the valid license plate information.

[0009] According to another aspect of the embodiments of this application, a tail license plate evidence collection device based on multi-source fusion is provided, the device comprising:

[0010] A first processing unit is configured to, in response to determining that a collision event has occurred, lock at least one responsible target from candidate targets continuously monitored by the vehicle's lidar prior to the collision; wherein the license plate of the responsible target is displayed at the rear.

[0011] The second processing unit is used to guide the vehicle-mounted camera to continuously track and image the location area indicated by the license plate location information of the responsible target provided by the lidar in real time, so as to obtain a license plate image sequence.

[0012] The third processing unit is used to identify and verify the tail license plate image sequence. If the valid license plate information of the responsible target is obtained before the responsible target escapes, then an evidence package uniquely corresponding to the responsible target is generated based on the valid license plate information.

[0013] According to another aspect of the embodiments of this application, an electronic device is provided, including: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus;

[0014] The memory is used to store at least one executable instruction that causes the processor to perform the operation of the multi-source fusion-based tail license plate evidence collection method as described above.

[0015] According to another aspect of the embodiments of this application, a computer-readable storage medium is provided, the storage medium storing at least one executable instruction, the executable instruction causing an electronic device / apparatus to perform the operation of the multi-source fusion-based tail license plate evidence collection method as described above.

[0016] The solution provided in this application continuously monitors candidate targets using LiDAR before a collision occurs, ensuring accurate identification of the single responsible party from multiple targets, achieving predictive identification of "whoever is hit is identified." After a collision, the real-time license plate location information provided by LiDAR guides the camera to perform continuous tracking imaging in short-exposure mode. This solves the problem of easily losing track of targets during high-speed movement and eliminates motion blur through short exposure, ensuring continuous acquisition of clear license plate images. Based on this, multi-frame recognition verification is performed on the acquired image sequence, effectively overcoming the misidentification problem of single-frame recognition under adverse conditions such as low light at night and overexposed taillights, significantly improving recognition accuracy. Finally, the identified valid license plate information is encrypted and encapsulated to generate a standardized evidence package uniquely corresponding to the responsible target. The entire process can complete a closed loop in a short time, ensuring full-process evidence collection from collision confirmation to evidence solidification within the escape window. Compared to the existing solution relying solely on rear cameras with a recognition rate of less than 30%, this solution achieves a qualitative leap in evidence collection success rate, recognition accuracy, and legal validity of evidence.

[0017] The above description is merely an overview of the technical solutions of the embodiments of this application. In order to better understand the technical means of the embodiments of this application and to implement them in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the embodiments of this application more obvious and understandable, specific implementation methods of this application are described below. Attached Figure Description

[0018] The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0019] Figure 1 A flowchart illustrating a method for obtaining evidence based on multi-source fusion for license plate identification, provided in an embodiment of this application;

[0020] Figure 2 An architecture diagram of a tail license plate evidence collection system based on multi-source fusion provided for embodiments of this application;

[0021] Figure 3 A flowchart illustrating another method for obtaining tail license plate evidence based on multi-source fusion provided in this application embodiment;

[0022] Figure 4 A schematic diagram of the structure of a tail license plate evidence collection device based on multi-source fusion provided in an embodiment of this application;

[0023] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0024] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein.

[0025] In the process of determining liability in traffic accidents, license plates are key evidence. However, in collision incidents, especially when one's own vehicle is rear-ended by another vehicle or when it is necessary to determine the liability of the following or side vehicle in other collision scenarios, the impact force at the moment of collision, changes in lighting (such as at night, backlighting, etc.), and the possible escape behavior of the liable party make it difficult for traditional camera devices to capture clear and effective images of the license plate in a timely manner.

[0026] In particular, current vehicle license plate standards allow motorcycles to display only rear license plates, meaning that 99% of the approximately 210 million two-wheeled vehicles nationwide lack front license plates. This renders existing traffic monitoring systems (such as forward-facing cameras) unable to capture these front license plates. In traffic accident scenarios, especially after collisions involving motorcycles used for delivery or courier services, the offending motorcycle often escapes at high speeds (approximately 43-72 km / h) by skidding. The effective evidence-gathering window for rear-view systems is typically very short (usually ≤250 milliseconds), and conventional panoramic monitoring or long-exposure modes are prone to failure due to motion blur or light interference. Therefore, overcoming adverse conditions such as motion blur and low-light conditions at night within the millisecond-level escape window to complete the entire evidence-gathering process from target identification to evidence preservation has become a pressing technical challenge.

[0027] Based on this, this application provides a method for obtaining license plate information based on multi-source fusion. After a collision event, the method uses LiDAR to accurately lock and track the position of the license plate of the responsible target, thereby guiding the camera to perform targeted focus tracking and capture. Combined with multi-frame recognition and verification, the method can efficiently and reliably obtain the license plate information and generate an evidence package before the responsible target escapes.

[0028] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0029] It should be noted that the execution entity of the tail license plate evidence collection method based on multi-source fusion provided in this application embodiment can be a tail license plate evidence collection device based on multi-source fusion. This device can be deployed on a controller in a vehicle with data processing and control functions, such as a vehicle control unit (VCU), central controller, intelligent driving domain controller, or other electronic devices, as part of the vehicle's active safety or driving record system. This application embodiment does not impose any limitations, and the method of this application can be implemented by software, hardware, or a combination of software and hardware.

[0030] Figure 1 This is a flowchart illustrating a multi-source fusion-based tail license plate evidence collection method provided in an embodiment of this application. This method can be executed by a multi-source fusion-based tail license plate evidence collection device. Figure 1 As shown, the method may include the following steps:

[0031] Step 110: In response to determining that a collision event has occurred, at least one responsible target is identified from the candidate targets continuously monitored by the vehicle's lidar before the collision; wherein the license plate of the responsible target is displayed at the rear.

[0032] For example, the collision event in this application mainly refers to a physical contact event between a vehicle and another vehicle. The vehicle can be a car or a motorcycle, and the other vehicle mainly refers to a motorcycle with its license plate attached to its rear, or a two-wheeled vehicle with its license plate attached to its rear. In the embodiments of this application, determining that a collision event has occurred can be achieved in various ways, such as by sensor detection, vehicle bus signal triggering, or user manual triggering, etc., and the embodiments of this application are not limited. In addition, the authenticity of the collision event can also be confirmed by multimodal fusion detection to distinguish it from non-collision interference such as speed bumps and potholes. For example, when the accelerometer detects a negative acceleration exceeding a threshold, and the lidar detects that the relative distance to a candidate target shrinks to zero or point cloud overlap and deformation occurs in a very short time (e.g., milliseconds), then it is comprehensively determined that a collision event has actually occurred. For example, when the Inertial Measurement Unit (IMU) detects an impact force greater than a preset impact force (e.g., 0.8g) for a certain preset duration (e.g., ≥12 milliseconds), and the LiDAR detects that the distance change between the vehicle and any candidate target is greater than a certain distance (e.g., >0.3m / 0.1s), the relative speed is less than a certain speed (e.g., ≤20m / s), and the point cloud intersection-over-union ratio (IoU) is greater than a certain value (e.g., >0.9), the collision event is determined to have actually occurred.

[0033] Candidate targets refer to potential interactive objects continuously tracked by LiDAR before a collision occurs, including but not limited to two-wheeled vehicles such as motorcycles, electric vehicles, and bicycles traveling behind. Responsible targets refer to the specific targets selected from the candidate targets that collided with the vehicle; these targets can be accurately identified through preset screening criteria. LiDAR is a three-dimensional sensor that perceives the surrounding environment by emitting laser beams and receiving reflected signals. In this application, the LiDAR is used at least to continuously monitor the areas behind and to the sides of the vehicle, providing precise information such as the target's position, outline, distance, and speed.

[0034] Before a collision occurs, the vehicle's LiDAR continuously operates, performing three-dimensional perception of targets within a pre-defined sector behind and to the sides of the vehicle (e.g., 30° directly behind, 20° to the left and right, and within a distance of 0-8 meters). The vehicle establishes a tracking record for each target entering the monitoring area, including its position coordinates, trajectory, speed, outline point cloud, vehicle type, and other information, forming a candidate target pool. When a collision event is determined, the responsible target locking logic is immediately activated, selecting one or more responsible targets strongly correlated with the collision from the candidate target pool based on the perception data at the moment of the collision.

[0035] This pre-collision monitoring method, which involves continuous monitoring before a collision and instant locking during the collision, not only can the responsible party be quickly identified at the moment of the collision, avoiding the time delay of temporary searches afterward, but even if the collision causes a drastic change in relative position, the responsible target can be locked based on historical trajectories to avoid loss.

[0036] Step 120: Based on the real-time tail license plate location information of the responsible target provided by the lidar, guide the vehicle-mounted camera to continuously track and image the location area indicated by the tail license plate location information to obtain a tail license plate image sequence.

[0037] For example, the tail license plate position information is the precise spatial position of the tail license plate area of ​​the responsible target, output in real time by the LiDAR. It can be represented as three-dimensional coordinates (x, y, z), or as distance, azimuth, and pitch angle relative to the vehicle. Considering the actual installation scenario, the tail license plate coordinates can be offset by the Z-axis (e.g., 0.2 meters) to adapt to the actual height of the license plate, based on the overall target coordinates.

[0038] Once the target is locked, the LiDAR can continuously output the location information of the license plate area of ​​the target vehicle at a high frame rate (e.g., 120fps, corresponding to approximately 8.33ms per frame). Then, the license plate location information provided by the LiDAR is used as a control command to drive the pan-tilt head of the vehicle-mounted camera to rotate to the corresponding position and adjust the camera's focal length, ensuring the lens focus remains locked on the location area indicated by the LiDAR for continuous shooting to obtain a sequence of license plate images. For example, the license plate detection device can map the license plate location information provided by the LiDAR to the image pixel coordinate system using a pre-calibrated coordinate transformation model (e.g., a transformation matrix based on a pinhole camera model). After transformation, a tracking sub-window (e.g., 800×400 pixels) centered on the mapping point is generated on the image plane. This sub-window represents the area including the license plate that needs to be focused on. Guided by the LiDAR, the vehicle-mounted camera continuously captures images of this specific coordinate area, with each frame aligned based on the latest LiDAR location information, creating a temporally continuous "focus tracking" effect, thus forming a sequence of license plate images with temporal continuity between image frames.

[0039] Optionally, to eliminate motion blur, the camera can be controlled to use a short exposure mode during shooting. Short exposure mode refers to the camera using an extremely short exposure time (e.g., 1 / 1000 second or even less) to capture images. In collision or escape scenarios, the relative speed of motion is extremely high, and long exposures will produce motion blur. Using short exposure mode can effectively "freeze" fast-moving vehicles, capturing clear license plate textures, thereby eliminating motion blur. Additionally, to compensate for insufficient light intake caused by short exposure, gain can be increased simultaneously (e.g., +6dB) to adapt to low-light scenarios such as nighttime. Furthermore, 64-zone ROI metering technology can be used to avoid overexposure of taillights interfering with the license plate area. Adjusting appropriate shooting parameters (such as gain, exposure time, focus mode, white balance, etc.) can further improve image sharpness and accuracy.

[0040] This application solves the problem of cameras being unable to continuously lock onto targets during high-speed relative motion by using laser real-time coordinate guidance and frame-level tracking imaging, providing a clear image foundation for subsequent recognition.

[0041] Step 130: Recognize and verify the tail license plate image sequence. If the valid license plate information of the responsible target is obtained before the responsible target escapes, then generate an evidence package that uniquely corresponds to the responsible target based on the valid license plate information.

[0042] For example, recognition verification refers to analyzing and processing the acquired image sequence to extract license plate characters, and verifying the accuracy of the recognition results through methods such as multi-frame comparison. Valid license plate information refers to information that can uniquely identify the license plate number and region of a motor vehicle. "Before the responsible target escapes" refers to the time window from the occurrence of the collision to the point where the responsible target is completely out of the vehicle's perception range; for high-speed two-wheeled vehicles escaping, this window is typically ≤250 milliseconds. All processing in this application must be completed within this window.

[0043] For example, the license plate identification device can perform optical character recognition (OCR) on the image acquired in step 120 in real time to obtain the license plate character recognition result for that frame, and simultaneously output the confidence score of the recognition result (e.g., 0-100%). If the recognition results of multiple consecutive frames (e.g., three frames) are completely identical, and the recognition confidence score of each frame reaches or exceeds a preset threshold (e.g., 95%), then the recognition result is confirmed as valid license plate information of the responsible target. Then, the license plate identification device can automatically encapsulate the data to generate an evidence package uniquely corresponding to the responsible target for subsequent legal tracing and evidence preservation. For example, multi-dimensional original evidence related to this collision event, including but not limited to: valid license plate information, raw frames (Raw images) from the tail plate image sequence, point cloud data of the responsible target acquired synchronously with the images, geographical location information at the time of the collision (such as GPS positioning), and precise timestamps synchronized via the Generalized Precision Time Protocol (GPTP), are encrypted using encryption algorithms such as the national cryptographic standard SM4 hash algorithm to generate a unique digital fingerprint for the evidence package. This standardized evidence package is then written into an immutable storage area (such as a secure encrypted storage chip), completing the entire evidence collection process. The entire process, from collision triggering to evidence storage completion, is controlled to be completed before the responsible target escapes. If valid license plate information of the responsible target cannot be obtained in time before the target escapes (e.g., the target has no license plate or recognition fails), other processing steps can be initiated (such as recording auxiliary features) to ensure no evidence is missing.

[0044] The tail license plate evidence collection method based on multi-source fusion provided in this application achieves zero-delay response of "whoever is hit is locked" through continuous monitoring by LiDAR before the collision, locking the responsible target at the moment of collision, saving the time of re-searching for the target in traditional solutions. After the collision, the tail license plate position information provided by LiDAR in real time guides the camera to continuously track and focus, realizing parallel processing of "positioning and capturing at the same time", ensuring that the capture process is synchronized with the target's escape, effectively avoiding losing track. At the same time, a pipeline operation mode of capturing, recognizing, and packaging at the same time is adopted. Once the image verification is passed, the evidence package is generated immediately, and the evidence is solidified before the target completely escapes. This series of interlocking collaborative mechanisms enables the entire process to be closed in a very short time, ensuring that the entire process of evidence collection from collision confirmation to evidence solidification is achieved within the escape window. Compared with the current situation where the recognition rate of the existing rear camera solution is less than 30%, this solution has achieved a qualitative leap in terms of evidence collection success rate, recognition accuracy, and legal effect of evidence.

[0045] For example, Figure 2This is an architecture diagram of a tail license plate evidence collection system based on multi-source fusion, provided for an embodiment of this application. Figure 2 As shown, the multi-source fusion-based tail license plate evidence collection system includes: a core processing chip FPGA SoC, an inertial measurement unit (IMU), a lidar, and a camera array (at least three 4K 120 fps high-definition cameras: a rear-facing camera and three side-facing cameras). The IMU collects data 1000 times per second to detect vehicle collision impact parameters. The lidar identifies the target's position, outline, and distance. The camera array, via the vehicle's Mobile Industry Processor Interface (MIPI), can synchronize with the lidar through microsecond-level triggering. After a collision, the corresponding camera is dynamically activated according to the impact direction, capturing images in a specific sub-window to save bandwidth and cover all scenarios, including side collisions and rear-end collisions. The core processing chip FPGA SoC is responsible for high-speed scheduling of sensor data and algorithm execution to realize the multi-source fusion-based tail license plate evidence collection method of this application.

[0046] Figure 3 This is a flowchart illustrating another method for obtaining license plate evidence based on multi-source fusion, provided in an embodiment of this application. This method can be executed by a multi-source fusion-based license plate evidence obtaining device, which can be integrated into a vehicle as part of its active safety or driving recorder system. Figure 3 As shown, the method may include the following steps:

[0047] Step 310: Determine whether a collision event actually occurred through multimodal fusion detection.

[0048] If so, proceed to step 320.

[0049] For example, multimodal fusion detection refers to a detection method that integrates data from multiple different types of sensors for comprehensive judgment. In this embodiment, the license plate evidence collection device can collect data from different sensors in real time, including but not limited to: acceleration signals from inertial measurement units, signals from dedicated collision sensors, point cloud data from lidar, and image data from vehicle-mounted cameras, etc., for multiple verifications. When a collision event is determined to have occurred based on multimodal data, it is considered that a real collision event has occurred. This application ensures through multi-sensor fusion that subsequent evidence collection processes are only executed when a real collision occurs, avoiding invalid triggering and false triggering.

[0050] Optionally, in one possible embodiment, determining a real collision event through multimodal fusion detection may include:

[0051] S1. Impact parameters are detected by inertial measurement unit, and distance changes, relative speeds, and point cloud intersection-to-union ratios between the vehicle and any candidate target are detected by lidar.

[0052] S2. If the impact parameters, distance change, relative velocity, and point cloud intersection ratio simultaneously meet the corresponding preset physical trigger thresholds, then wake up the camera corresponding to the collision direction to capture the first frame image.

[0053] S3. Detect the first frame image using a visual algorithm. If real physical contact is detected, then a collision event is confirmed to have actually occurred.

[0054] For example, the inertial measurement unit and lidar on the vehicle can be set to a normally-on state. The inertial measurement unit detects the impact parameters of the vehicle in real time at a high-frequency sampling rate of 1kHz. The impact parameters include at least the magnitude of the impact force and the corresponding duration. At the same time, the lidar detects the relative motion parameters between the vehicle and any candidate target within the surrounding monitoring range. The relative motion parameters include at least the distance change (ΔR), relative velocity (v), and intersection-over-union ratio (IoU) of the point cloud.

[0055] Then, it is determined whether the parameters detected in S1 simultaneously meet the corresponding preset physical trigger thresholds. Specifically, when the impact force is greater than a preset value (e.g., 0.8g) and the duration is greater than or equal to a preset duration (e.g., 12 milliseconds), it can effectively filter out common interferences such as speed bumps and potholes; if the change in distance between the vehicle and any candidate target is greater than a preset change (e.g., 0.3 meters / 0.1 seconds), it indicates that a target is rapidly approaching; if the relative speed between the vehicle and the candidate target is less than or equal to a preset speed (e.g., 20 m / s), it corresponds to the typical escape speed range of a motorcycle; if the intersection-union ratio of the point cloud between the vehicle and the candidate target is greater than a preset intersection-union ratio (e.g., 0.9), it indicates that the outlines of the two are highly overlapping, and physical contact has occurred or is about to occur. When the above impact parameters, distance change, relative speed, and point cloud intersection-union ratio simultaneously meet their respective preset physical trigger thresholds, it is determined that the physical level has collision characteristics, and a first trigger signal is generated. According to the collision direction indicated in the impact parameters (e.g., forward or sideways), the camera (rear or side-mounted) corresponding to that direction is activated to acquire the first frame image.

[0056] Furthermore, a visual AI algorithm analyzes the overlapping areas of the target's contours, pixel grayscale abrupt changes, and texture changes in the contact area in the first frame image captured by the camera awakened in S2 to determine whether there are real physical contact traces. If real physical contact is detected, a second trigger signal is generated to ultimately confirm that a collision event has actually occurred, and the subsequent responsible target locking and evidence collection process is allowed; if no physical contact traces are detected, it is determined to be a false trigger, and the subsequent process is terminated to avoid invalid startup due to a non-collision event.

[0057] This optional embodiment uses the mechanical parameters detected by the inertial measurement unit (IMU) and the kinematic parameters detected by the lidar to form a dual physical triggering condition, which is then superimposed with visual image verification, forming a triple verification mechanism from the physical to the visual level. This ensures that the evidence collection process is only initiated when a real collision occurs, fundamentally avoiding false triggering. Furthermore, waking the camera to acquire the first frame image only after the dual physical triggering conditions are met avoids power waste and data redundancy caused by the camera being in continuous operation. Throughout the detection process, the gPTP protocol ensures that the timestamp synchronization error of the IMU, lidar, and camera is less than 200 nanoseconds, thus laying a solid foundation for the subsequent accurate mapping of laser coordinates and image pixels.

[0058] Step 320: From the candidate targets continuously monitored by the vehicle's lidar before the collision, identify at least one responsible target; wherein the license plate of the responsible target is displayed at the rear.

[0059] It should be noted that the specific implementation of step 320 can be referred to the description of step 110, and will not be repeated here.

[0060] Optionally, in one possible embodiment, identifying at least one responsible target from candidate targets continuously monitored by the vehicle's lidar before the collision may include:

[0061] S10. Before a collision occurs, the vehicle continuously monitors targets in the preset sectors behind and to the side of the vehicle using its own LiDAR, establishes a candidate target pool, and records the movement trajectory and vehicle model information of each candidate target.

[0062] S20. When a collision occurs, at least one responsible target is locked from the candidate target pool according to preset screening criteria; wherein, the preset screening criteria include spatial proximity criterion, trajectory intersection criterion and vehicle model matching criterion.

[0063] For example, the lidar, as the vehicle's primary environmental perception sensor, leverages its wide field of view and high ranging accuracy to pre-monitor specific areas around the vehicle. Considering that license plate detection primarily targets two-wheeled vehicles that may collide with the vehicle from the rear or side, the monitoring focus is placed on preset sectors behind and to the sides of the vehicle. For instance, these preset sectors could be a range of 30° directly behind the vehicle, 20° to the left and right, and 0-8 meters. The specific angles and distances can be dynamically calibrated based on the lidar's field of view (FOV) and the actual road scene; this embodiment does not impose any limitations.

[0064] LiDAR continuously emits laser beams and receives echoes, generating 3D point cloud data. Through point cloud clustering, target detection, and tracking algorithms, stable groups of reflective points in continuous point cloud frames are identified as individual "targets." Each identified target confirmed to have entered a preset sector is added to a candidate target pool. This candidate target pool is a dynamic data structure used to temporarily store information about all potentially relevant targets. The license plate verification device continuously tracks each candidate target in the pool, recording its historical "trajectory" (including temporal data such as position, speed, acceleration, and heading angle) and identifying its "vehicle type information" (e.g., identifying it as a sedan, motorcycle, truck, or bus through point cloud contour matching). This pre-monitoring and recording mechanism accumulates rich foundational data for rapid and accurate liability determination after a collision.

[0065] Furthermore, when a collision is confirmed through multimodal fusion detection, the license plate evidence collection device immediately initiates the responsibility target locking process. It quickly queries the "candidate target pool" established before the collision and rapidly matches the candidate targets in the pool according to preset screening criteria to identify the most likely party responsible for the accident. The preset screening criteria are a multi-dimensional comprehensive judgment logic, including: spatial proximity criterion, trajectory intersection criterion, and vehicle model matching criterion.

[0066] Specifically, the spatial distance between each target in the candidate target pool and the collision point of the vehicle can be calculated. The collision point can be estimated by fusing the IMU impact direction, the overlapping area of ​​the LiDAR point cloud, and the vehicle dynamics model. If the spatial distance between the candidate target and the collision point is less than a preset distance (e.g., 0.5 meters), it can be considered that the target is in physical contact, rather than a vehicle traveling in parallel lanes. At the same time, the historical motion trajectory of the candidate target is compared with the motion trajectory of the vehicle before the collision, and the trajectory overlap is calculated. If the overlap between the motion trajectory of the candidate target and the motion trajectory of the vehicle exceeds a preset overlap (e.g., 90%), it is considered a target whose motion inevitably intersects with the vehicle's motion. In addition, the recorded vehicle type identification confidence score of each candidate target is called to determine whether its vehicle type matching degree exceeds a preset confidence score (e.g., motorcycle confidence score > 0.8) to determine that it is a two-wheeled vehicle and exclude pedestrians, four-wheeled vehicles, and other objects. For example, only targets that simultaneously meet the criteria of "spatial distance < 0.5 meters", "trajectory overlap > 90%", and "vehicle type confidence score > 0.8" are identified as responsible targets.

[0067] In most scenarios, the aforementioned triple screening process can identify a single responsible target. In rare special cases where multiple targets simultaneously meet the criteria, the single responsible target can be further determined based on the principle of closest spatial distance or the principle of maximum intersection-union ratio, and a unique responsibility ID can be assigned to it. This responsibility ID will serve as a unique identifier throughout all subsequent processing stages, used to associate all images, point clouds, recognition results, and evidence data related to that target.

[0068] This optional embodiment, through continuous pre-collision monitoring and the construction of a candidate target pool, advances target identification and tracking. This eliminates the need to start searching for targets from scratch upon collision; instead, it directly filters from the established candidate pool, saving valuable time for subsequent closed-loop processes involving collecting valid license plate information before the responsible party escapes. Simultaneously, the triple-screening principle ensures the uniqueness of the responsible party, guaranteeing accurate identification of the true party responsible for the collision in complex multi-target scenarios and preventing the accidental locking of parallel vehicles in adjacent lanes.

[0069] Step 330: Map the tail license plate location information of the responsible target provided by the LiDAR in real time to the image pixel coordinate system through a coordinate transformation algorithm, and generate and dynamically update the position of the tracking sub-window.

[0070] For example, after locking onto the target, the lidar can be controlled to output the location information of the license plate area at the rear of the vehicle at a high frame rate in real time, such as the three-dimensional coordinates (x, y, z) of the center point of the license plate. Then, the license plate identification device can utilize pre-calibrated joint extrinsic parameters between the lidar and the camera, as well as the camera's intrinsic parameter matrix, to accurately map the license plate location information onto pixel coordinates on a two-dimensional image plane using a projection equation. Centered on the transformed pixel coordinates, a tracking sub-window of a specific size (e.g., 800×400 pixels) can be generated. The specific size can be dynamically adjusted according to the target distance; for example, the closer the distance, the larger the window. This embodiment does not impose limitations. This sub-window represents the area that needs to be focused on in the current frame. In the next frame, as the target moves, the lidar provides new license plate location coordinates, and the coordinate transformation is re-executed to generate a new sub-window position, achieving "frame-level dynamic updates." This ensures that the sub-window always follows the target's movement, surrounding the license plate and achieving true "focus tracking."

[0071] Step 340: Control the camera corresponding to the collision direction to continuously capture images of the tracking sub-window in short exposure mode to obtain the tail license plate image sequence.

[0072] For example, during the aforementioned process of determining whether a collision has actually occurred, the license plate evidence collection device will automatically activate the corresponding camera based on the direction of the collision. For instance, if a frontal impact is detected, the rear-view camera is activated; if a left-side impact is detected, the left-side camera is activated. This mechanism ensures that the camera with the best viewing angle can be used to capture images regardless of the direction of the collision, avoiding blind spots. Simultaneously, to avoid image blurring caused by the high-speed movement of the fleeing target or the impact of the collision, the camera will use a short-exposure mode (e.g., a shutter speed higher than 1 / 1000 of a second) to freeze the fast-moving vehicle and eliminate image motion blur caused by relative motion. The camera does not capture the entire large image but focuses on continuously capturing images of the tracking sub-window area determined in step 330, thereby reducing the data transmission and processing burden while obtaining a continuous multi-frame image sequence focused on the license plate area. This "point-and-shoot" strategy greatly improves the efficiency of collecting effective data.

[0073] Step 350: Perform optical character recognition on the images in the tail plate image sequence to obtain the recognition result and corresponding confidence level of each frame image.

[0074] For example, after images are acquired, the acquired image frames are sequentially input into an Optical Character Recognition (OCR) engine. This OCR engine, based on a deep learning model, can recognize Chinese characters, letters, numbers, and special characters in license plates. For each image frame, the OCR engine not only outputs the recognized license plate number but also a "confidence" score (usually a decimal between 0 and 1). This score reflects the model's confidence in the recognition result. A higher confidence score means a clearer image, more complete characters, and a more reliable recognition result.

[0075] Optionally, in one possible embodiment, in order to improve the accuracy of image recognition, before performing optical character recognition on the images in the tail license plate image sequence, super-resolution enhancement processing can be performed on the images in the tail license plate image sequence, and then the enhanced images can be recognized.

[0076] For example, due to factors such as shooting distance and camera resolution, the captured license plate image may have insufficient resolution and blurry character details. To improve image quality, the license plate identification device can also process each captured frame of the license plate image in real time using the EDSR×2 algorithm, doubling the image resolution. The EDSR (Enhanced Deep Super-Resolution) algorithm is an advanced super-resolution algorithm that learns the mapping relationship between low-resolution and high-resolution images through a deep neural network, effectively restoring the edge and texture details of the license plate characters, making the originally blurry characters clearly discernible.

[0077] In this embodiment, super-resolution processing is a secondary enhancement performed at the software level, which can complement the short exposure at the hardware level. Short exposure solves motion blur, while software super-resolution solves insufficient resolution. The combination of the two provides high-quality input images for subsequent optical character recognition (OCR).

[0078] Step 360: Before the responsible target escapes, when the recognition results of multiple consecutive frames are completely identical and the confidence level of each frame is greater than or equal to the preset threshold, the recognition result is confirmed as the valid license plate information of the responsible target, and an evidence package uniquely corresponding to the responsible target is generated based on the valid license plate information.

[0079] For example, to eliminate accidental errors that may occur in single-frame recognition, the license plate identification device also pre-sets a confidence threshold (e.g., 0.85) and a consecutive frame requirement (e.g., 3 consecutive frames). Only when the OCR recognition result strings of multiple consecutive frames (e.g., the Nth frame, N+1th frame, N+2th frame) are completely identical, and the confidence level of each frame is greater than or equal to the preset threshold, is the recognition result determined to be valid license plate information. This multi-frame verification mechanism effectively eliminates misidentification caused by momentary occlusion, reflection, or motion blur, ensuring that the final obtained license plate information is highly reliable. If valid license plate information of the responsible party is obtained before they escape, an evidence package uniquely corresponding to the responsible party is generated based on the valid license plate information for subsequent legal tracing and evidence preservation.

[0080] Optionally, in one possible embodiment, generating an evidence package uniquely corresponding to the liability target based on valid license plate information may include:

[0081] Multiple pieces of evidence, including at least valid license plate information, collected raw image data, synchronously collected laser point cloud data, location data, and timestamps, are processed by an encryption algorithm and packaged into an unalterable evidence package that uniquely corresponds to the responsible target.

[0082] For example, upon obtaining reliable license plate information, the evidence preservation process is initiated immediately. First, all original evidence related to the collision event is collected, including: the aforementioned identified valid license plate information, the collected original image data, the laser point cloud data of the responsible target collected synchronously with the images, the vehicle's current GPS / BeiDou coordinates, and the precise time information synchronized with the time. Then, a hash operation (such as SHA-256) is performed on the entire evidence package to generate a unique digital digest (hash value); this digest is then encrypted using the vehicle's or platform's private key to form a digital signature. Finally, the original evidence data, digital signature, and digital certificate are packaged into a structured file and written as an evidence package to an immutable storage area (e.g., a secure encrypted storage chip), completing the entire evidence collection process.

[0083] Optionally, in one possible embodiment, the method provided in this application may further include: if no valid license plate information is identified in the image captured by the vehicle-mounted camera corresponding to the collision direction, then switch to a vehicle-mounted camera at another angle for re-capture; if no valid license plate information is obtained after re-capture, then record the vehicle model characteristics, movement trajectory, and current location information of the responsible target, and include them in the evidence package as auxiliary evidence.

[0084] For example, if it is determined that valid license plate information cannot be obtained based on the image captured by the camera corresponding to the collision direction (e.g., the recognition confidence is lower than a preset threshold for N consecutive frames, or the OCR engine output is empty), a re-shooting process can be triggered. Specifically, the license plate evidence collection device can automatically switch to the vehicle's onboard camera at other angles, such as the corresponding viewpoint from the side-view camera, rear-view camera, or surround-view camera adjacent to the collision direction. The switching can be based on the real-time position information and movement trend of the responsible target in the LiDAR point cloud, selecting the best-view camera that can cover the license plate area of ​​the responsible target's rear. This camera also uses short-exposure mode to continuously capture images of the LiDAR-guided tracking sub-window and repeats the multi-frame recognition verification process from steps 350 to 360 to attempt to re-acquire valid license plate information. If valid license plate information is still not obtained after re-shooting (e.g., the target license plate is severely obscured, damaged, or completely out of the view of all cameras), auxiliary evidence of the responsible target can be automatically recorded, such as vehicle characteristics like body color and wheel style, movement trajectory (e.g., speed, acceleration, escape direction), and current location information (e.g., GNSS positioning). This auxiliary evidence, after being collected, is included in the evidence package along with existing raw image data, laser point cloud data, and timestamps. Even if the exact license plate cannot be obtained, this auxiliary evidence can provide crucial clues for subsequent manual investigation, vehicle tracking, and liability determination.

[0085] This optional embodiment significantly improves the system's robustness in complex environments by introducing a multi-camera switching and reshooting mechanism, overcoming the problem of evidence collection failure caused by obstruction or poor angles from a single perspective. Furthermore, even if the license plate text is ultimately not obtained, the recorded vehicle characteristics, movement trajectory, and location information, among other auxiliary evidence, provide valuable tracing evidence for the accident investigation, preserving as much relevant information as possible and enhancing the integrity of the evidence chain.

[0086] For example, to more intuitively demonstrate the millisecond-level closed-loop capability of this solution, Table 1 below summarizes the typical time consumption of each step when the camera is working at 120fps.

[0087] Table 1

[0088]

[0089] As can be seen from the timing in Table 1 above, based on the sequential operation mode of simultaneous capture, identification, and confirmation in this application, as well as the interlocking processing flow, the entire process from collision confirmation to evidence solidification can be completed within 48ms, which is much smaller than the 250ms escape window, thus enabling evidence collection before escape.

[0090] The tail license plate evidence collection method based on multi-source fusion provided in this application provides real-time confirmation of collision events through multi-modal fusion detection, ensuring zero-delay response for evidence collection initiation. It utilizes LiDAR for continuous monitoring before collision to achieve predictive locking of the responsible target, and relies on coordinate transformation algorithms to map the tail license plate location information to the image coordinate system in real time to dynamically generate a tracking sub-window. This enables parallel processing of positioning and capturing during the escape process, avoiding loss of tracking and time due to target movement or occlusion. Furthermore, it performs complete consistency verification and confidence threshold judgment on the collected tail license plate image sequence through multi-frame recognition results. A streamlined operation mode of simultaneous capturing, recognition, and confirmation is adopted to quickly eliminate single-frame misidentification and random interference, achieving high-accuracy license plate information extraction before the target completely escapes. Finally, the valid license plate information, along with the original image, LiDAR point cloud, positioning data, and timestamps, are encrypted and encapsulated into an immutable evidence package, ensuring the integrity and legal validity of the evidence from the data source. Thus, the aforementioned interconnected mechanism enables the entire process to be completed efficiently before the escape window closes, realizing full-process evidence collection from collision confirmation to evidence solidification, effectively improving the success rate of evidence collection, the accuracy of identification, and the legal effect of evidence.

[0091] Figure 4 This is a schematic diagram of a tail-plate evidence collection device based on multi-source fusion, provided as an embodiment of this application. Figure 4 As shown, the device 40 includes a first processing unit 401, a second processing unit 402, and a third processing unit 403.

[0092] The first processing unit 401 is configured to, in response to determining that a collision event has occurred, lock at least one responsible target from candidate targets continuously monitored by the vehicle's lidar before the collision; wherein the license plate of the responsible target is displayed at the rear.

[0093] The second processing unit 402 is used to guide the vehicle-mounted camera to continuously track and image the location area indicated by the license plate location information of the responsible target provided by the lidar in real time, so as to obtain a license plate image sequence.

[0094] The third processing unit 403 is used to identify and verify the tail license plate image sequence. If the valid license plate information of the responsible target is obtained before the responsible target escapes, an evidence package uniquely corresponding to the responsible target is generated based on the valid license plate information.

[0095] In one alternative embodiment, the first processing unit 401 is specifically used for:

[0096] The actual collision events were determined by multimodal fusion detection.

[0097] In one alternative embodiment, the first processing unit 401 is specifically used for:

[0098] Impact parameters are detected by inertial measurement unit, and distance changes, relative speeds, and point cloud intersection-to-union ratios between the vehicle and any candidate target are detected by lidar.

[0099] If the impact parameters, distance change, relative velocity, and point cloud intersection ratio simultaneously meet the corresponding preset physical trigger thresholds, then the camera corresponding to the collision direction will be activated to capture the first frame image.

[0100] The first frame of the image is detected using a visual algorithm. If real physical contact is detected, it is determined that a collision event has actually occurred.

[0101] In one alternative embodiment, the first processing unit 401 is specifically used for:

[0102] Before a collision occurs, the vehicle's lidar continuously monitors targets within a preset sector behind and to the side of the vehicle, establishes a candidate target pool, and records the movement trajectory and vehicle model information of each candidate target.

[0103] When a collision occurs, at least one responsible target is identified from the candidate target pool according to preset screening criteria; the preset screening criteria include spatial proximity criterion, trajectory intersection criterion, and vehicle model matching criterion.

[0104] In one alternative embodiment, the second processing unit 402 is specifically used for:

[0105] The real-time tail license plate location information of the target provided by the LiDAR is mapped to the image pixel coordinate system through a coordinate transformation algorithm to generate and dynamically update the position of the tracking sub-window;

[0106] The camera corresponding to the direction of collision is controlled to continuously capture images of the tracking sub-window in short exposure mode to obtain the image sequence of the last license plate.

[0107] In one alternative embodiment, the third processing unit 403 is specifically used for:

[0108] Optical character recognition is performed on the images in the last plate image sequence to obtain the recognition result and corresponding confidence level of each frame image;

[0109] Before the responsible target escapes, when the recognition results of multiple consecutive frames are completely identical and the confidence level of each frame is greater than or equal to a preset threshold, the recognition result is confirmed as the valid license plate information of the responsible target, and an evidence package uniquely corresponding to the responsible target is generated based on the valid license plate information.

[0110] In one alternative embodiment, before performing optical character recognition on the images in the tail plate image sequence, the third processing unit 403 is further configured to:

[0111] Super-resolution enhancement processing is performed on the images in the last plate image sequence.

[0112] In one alternative embodiment, the third processing unit 403 is specifically used for:

[0113] Multiple pieces of evidence, including at least valid license plate information, collected raw image data, synchronously collected laser point cloud data, location data, and timestamps, are processed by an encryption algorithm and packaged into an unalterable evidence package that uniquely corresponds to the responsible target.

[0114] In one alternative embodiment, the second processing unit 402 is further configured to:

[0115] If no valid license plate information is identified in the image captured by the vehicle camera corresponding to the direction of the collision, the camera will be switched to another angle for reshooting.

[0116] If valid license plate information is still not obtained after re-photographing, the vehicle characteristics, movement trajectory, and current location information of the responsible party will be recorded and included in the evidence package as supplementary evidence.

[0117] As can be seen from the above, the tail license plate evidence collection device based on multi-source fusion provided in this application embodiment achieves zero-delay response through laser radar prediction and locking. After collision, the camera is guided to track and capture images in parallel with real-time coordinates. It adopts a pipeline operation mode of capturing, recognizing, and encapsulating images simultaneously, thereby quickly completing the entire closed loop from collision confirmation to evidence solidification within the escape window, which also effectively improves the success rate of evidence collection and the accuracy of recognition.

[0118] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The specific embodiments of this application do not limit the specific implementation of the electronic device.

[0119] like Figure 5 As shown, the electronic device may include: a processor 502, a communications interface 504, a memory 506, and a communications bus 508.

[0120] The processor 502, communication interface 504, and memory 506 communicate with each other via communication bus 508. Communication interface 504 is used to communicate with other network elements such as clients or other servers. The processor 502 executes program 510, specifically performing the relevant steps described in the embodiment of the multi-source fusion-based tail-plate evidence collection method.

[0121] Specifically, program 510 may include program code, which includes computer-executable instructions.

[0122] Processor 502 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. The electronic device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or they may be processors of different types, such as one or more CPUs and one or more ASICs.

[0123] Memory 506 is used to store program 510. Memory 506 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0124] Specifically, program 510 can be called by processor 502 to cause the electronic device to execute the relevant steps in the above embodiment of the tail license plate evidence collection method based on multi-source fusion.

[0125] This application provides a computer-readable storage medium storing at least one executable instruction. When the executable instruction is executed on an electronic device, it causes the electronic device to perform the multi-source fusion-based tail license plate evidence collection method in any of the above method embodiments.

[0126] The algorithms or displays provided herein are not inherently related to any particular computer, virtual system, or other device. Furthermore, the embodiments in this application are not directed to any particular programming language.

[0127] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of this application may be practiced without these specific details. Similarly, for the purpose of simplification and aiding understanding of one or more aspects of the invention, in the above description of exemplary embodiments of this application, various features of the embodiments are sometimes grouped together in a single embodiment, figure, or description thereof. The claims, which follow the detailed description, are hereby expressly incorporated into that detailed description, wherein each claim itself is a separate embodiment of this application.

[0128] Those skilled in the art will understand that the modules in the device of the embodiment can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiment can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components, except that at least some of such features and / or processes or units are mutually exclusive.

[0129] It should be noted that the above embodiments are illustrative of this application and not restrictive, and those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. This application can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be construed as limiting the order of execution.

Claims

1. A method for obtaining evidence based on the tail license plate number using multi-source fusion, characterized in that, The method includes: In response to determining that a collision event has occurred, at least one responsible target is identified from candidate targets continuously monitored by the vehicle's lidar prior to the collision; wherein the license plate of the responsible target is displayed at the rear. Based on the real-time tail license plate location information of the responsible target provided by the lidar, the vehicle-mounted camera is guided to continuously track and image the location area indicated by the tail license plate location information to obtain a tail license plate image sequence. The image sequence of the last license plate is identified and verified. If the valid license plate information of the responsible target is obtained before the responsible target escapes, then an evidence package uniquely corresponding to the responsible target is generated based on the valid license plate information.

2. The method according to claim 1, characterized in that, The determination of a collision event includes: The actual collision events were determined by multimodal fusion detection.

3. The method according to claim 2, characterized in that, The process of determining actual collision events through multimodal fusion detection includes: Impact parameters are detected by inertial measurement unit, and distance changes, relative speeds, and point cloud intersection-to-union ratios between the vehicle and any candidate target are detected by lidar. If the impact parameters, the distance change, the relative velocity, and the point cloud intersection-union ratio simultaneously meet the corresponding preset physical trigger thresholds, then the camera corresponding to the collision direction is activated to capture the first frame image. The first frame image is detected using a visual algorithm. If real physical contact is detected, it is determined that a collision event has actually occurred.

4. The method according to claim 1, characterized in that, The process of identifying at least one responsible target from candidate targets continuously monitored by the vehicle's lidar before the collision includes: Before a collision occurs, the vehicle's lidar continuously monitors targets within a preset sector behind and to the side of the vehicle, establishes a candidate target pool, and records the movement trajectory and vehicle model information of each candidate target. When a collision occurs, at least one responsible target is identified from the candidate target pool according to preset screening criteria; wherein, the preset screening criteria include spatial proximity criterion, trajectory intersection criterion, and vehicle model matching criterion.

5. The method according to claim 1, characterized in that, The step of guiding the vehicle-mounted camera to continuously track and image the location area indicated by the license plate location information of the responsible target in real time, based on the real-time license plate location information provided by the lidar, to obtain a license plate image sequence includes: The tail plate location information of the target provided in real time by the lidar is mapped to the image pixel coordinate system through a coordinate transformation algorithm to generate and dynamically update the position of the tracking sub-window; The camera corresponding to the collision direction is controlled to continuously capture images of the tracking sub-window in short exposure mode to obtain the tail license plate image sequence.

6. The method according to claim 1, characterized in that, The process of identifying and verifying the tail license plate image sequence involves, if, before the responsible party flees, obtaining valid license plate information of the target, then generating an evidence package uniquely corresponding to the responsible party based on the valid license plate information, including: Optical character recognition is performed on the images in the tail plate image sequence to obtain the recognition result and corresponding confidence level of each frame image; Before the responsible target escapes, when the recognition results of multiple consecutive frames are completely identical and the confidence level of each frame is greater than or equal to a preset threshold, the recognition results are confirmed as the valid license plate information of the responsible target, and an evidence package uniquely corresponding to the responsible target is generated based on the valid license plate information.

7. The method according to claim 5, characterized in that, Before performing optical character recognition on the images in the tail plate image sequence, the method further includes: Super-resolution enhancement processing is performed on the images in the tail plate image sequence.

8. The method according to any one of claims 1-7, characterized in that, The step of generating an evidence package uniquely corresponding to the responsible target based on the valid license plate information includes: Multiple pieces of evidence, including at least the valid license plate information, the collected raw image data, the synchronously collected laser point cloud data, the positioning data, and the timestamp, are processed by an encryption algorithm and then packaged into an unalterable evidence package that uniquely corresponds to the responsible target.

9. The method according to any one of claims 1-7, characterized in that, The method further includes: If no valid license plate information is identified in the image captured by the vehicle camera corresponding to the direction of the collision, the camera will be switched to another angle for reshooting. If valid license plate information is still not obtained after re-photographing, the vehicle characteristics, movement trajectory, and current location information of the responsible target will be recorded and included in the evidence package as supplementary evidence.

10. An electronic device, characterized in that, include: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction that causes the processor to perform the operation of the tail license plate evidence collection method based on multi-source fusion as described in any one of claims 1-9.

Images