Vehicle collision test method and system, electronic equipment and storage medium

Abstract

The invention relates to a vehicle collision test method and system, electronic equipment and a storage medium. The method comprises the steps that the method is applied to a collision test system, the collision test system is in communication connection with a test vehicle, a collision test task is received, and the collision test task comprises a target collision point arranged on a collision test runway, first visual image data collected based on the test vehicle, corresponding first inertial navigation data are corrected, and a first collision pose is obtained; and when it is detected that the test vehicle enters a preset key collision area in a first collision pose, correcting the corresponding second inertial navigation data based on second visual image data collected by the test vehicle to obtain a second collision pose, and predicting a target collision pose of the test vehicle according to the second collision pose. The target collision pose is used for controlling the test vehicle to collide with the target collision point. According to the scheme, the test vehicle can be accurately controlled to collide with the target collision point, and it is ensured that the test vehicle efficiently completes the collision test task.

Description

Technical Field

[0001] The present application relates to the field of vehicle testing technology, and in particular to a vehicle collision testing method, system, electronic device, and storage medium. Background Art

[0002] With the continuous development of intelligent driving technology, Autonomous Emergency Braking (AEB) has become an important component in improving vehicle safety and reducing collision risks. When a potential collision risk is identified, the AEB system can proactively implement braking intervention, thereby reducing the probability of a collision.

[0003] In practical applications, vehicles used in AEB testing scenarios are typically equipped with a driving robot or autonomous driving platform to precisely control the steering wheel, accelerator, and brakes to achieve repeatable and controllable collision maneuvers. However, relevant collision testing technologies generally suffer from insufficient positioning accuracy, making it difficult to meet the requirements of AEB performance evaluation. Summary of the Invention

[0004] In order to solve or partially solve the problems existing in the related art, the present application provides a vehicle collision test method, system, electronic equipment and storage medium, which can significantly improve the positioning accuracy of the collision test and efficiently complete the collision test task.

[0005] In a first aspect, the present application provides a vehicle collision test method, which is applied to a collision test system, wherein the collision test system is communicatively connected to a test vehicle, and the method comprises: receiving a collision test task, wherein the collision test task includes a target collision point set on a collision test track; Based on the first visual image data collected by the test vehicle, correcting the corresponding first inertial navigation data to obtain a first collision posture; When detecting that the test vehicle enters a preset critical collision area in the first collision posture, correcting corresponding second inertial navigation data based on second visual image data collected by the test vehicle to obtain a second collision posture; A target collision posture of the test vehicle is predicted based on the second collision posture, and the target collision posture is used to control the test vehicle to collide with the target collision point.

[0006] In one embodiment, the correcting corresponding first inertial navigation data based on the first visual image data collected by the test vehicle to obtain the first collision posture includes: Acquiring first visual image data and first inertial navigation data collected when the test vehicle travels along the collision test track; Performing feature extraction on the first visual image data to obtain a first visual feature point and first visual feature data of the first visual feature point; The first visual feature data is used to correct the corresponding first inertial navigation data to obtain the first collision posture.

[0007] In one example, the using the first visual feature data to correct the corresponding first inertial navigation data to obtain the first collision posture includes: Extracting a first acquisition time point and a first feature point deviation posture of the first visual feature point from the first visual feature data; Extracting a first measurement pose corresponding to the first acquisition time point from the first inertial navigation data; The first measured posture is corrected using the first feature point deviation posture to obtain the first collision posture.

[0008] In one example, when detecting that the test vehicle enters a preset critical collision area in the first collision posture, correcting corresponding second inertial navigation data based on second visual image data collected by the test vehicle to obtain a second collision posture includes: Acquiring second visual image data of the test vehicle traveling in the first collision posture; If a second visual feature point is identified from the second visual image data and the collision distance between the test vehicle and the target collision point is less than a set distance threshold, determining that the test vehicle has entered the critical collision area; Acquire second visual feature data corresponding to the second visual feature point; The second visual feature data is used to correct the corresponding second inertial navigation data to obtain the second collision posture.

[0009] In one example, the using the second visual feature data to correct the corresponding second inertial navigation data to obtain the second collision posture includes: Extracting a second acquisition time point and a second feature point deviation posture of the second visual feature point from the second visual feature data; Extracting a second measurement pose corresponding to the second acquisition time point from the second inertial navigation data; The second measured posture is corrected using the second feature point deviation posture to obtain the second collision posture.

[0010] In one example, predicting a target collision posture of the test vehicle based on the second collision posture includes: Acquiring target visual image data and target inertial navigation data of the test vehicle traveling in the second collision posture; performing data preprocessing on the target visual image data and the target inertial navigation data respectively to generate model input data; The model input data is input into a preset posture prediction model to output the target collision posture.

[0011] In one example, the posture prediction model is constructed as follows: Constructing an initial neural network model and a training sample set; wherein the training sample set includes at least historical inertial navigation data, historical visual image data, and historical real poses; Using the training sample set to train the initial neural network model to obtain a predicted pose; The training error between the predicted posture and the historical true posture is calculated. When the training error converges to a preset threshold, the model training is completed and the posture prediction model is generated.

[0012] A second aspect of the present application provides a collision test system, wherein the collision test system is communicatively connected to a test vehicle; wherein the system comprises: A collision test task receiving module, configured to receive a collision test task, wherein the collision test task includes a target collision point located on a collision test track; a first posture correction module, configured to correct corresponding first inertial navigation data based on first visual image data collected by the test vehicle to obtain a first collision posture; a second posture correction module, configured to correct corresponding second inertial navigation data based on second visual image data collected by the test vehicle to obtain a second collision posture when detecting that the test vehicle enters a preset critical collision area in the first collision posture; A target collision posture prediction module is used to predict a target collision posture of the test vehicle according to the second collision posture, and the target collision posture is used to control the test vehicle to collide with the target collision point.

[0013] A third aspect of the present application provides an electronic device, including: processor; and The memory stores executable codes thereon, and when the executable codes are executed by the processor, the processor is caused to execute the method described above.

[0014] A fourth aspect of the present application provides a computer-readable storage medium having executable code stored thereon. When the executable code is executed by a processor of an electronic device, the processor is caused to execute the method described above.

[0015] A fifth aspect of the present application provides a computer program product, which includes computer instructions, and when the computer instructions are executed by a processor, implements the method described above.

[0016] The technical solution provided by this application may include the following beneficial results: The technical solution of the present application is applied to a collision test system, which is communicatively connected to a test vehicle and receives a collision test task. The collision test task includes a target collision point set on a collision test track. Based on first visual image data collected by the test vehicle, the corresponding first inertial navigation data is corrected to obtain a first collision posture. When it is detected that the test vehicle enters a preset critical collision area with the first collision posture, the corresponding second inertial navigation data is corrected based on second visual image data collected by the test vehicle to obtain a second collision posture. According to the second collision posture, the target collision posture of the test vehicle is predicted, and the target collision posture is used to control the test vehicle to collide with the target collision point.

[0017] Compared to related technologies, this application achieves multi-stage pose correction during the collision test process by combining visual image data collected by the test vehicle with inertial navigation data, effectively improving positioning accuracy. Furthermore, when the test vehicle enters the critical collision zone, the target collision pose is further predicted based on the corrected collision pose, thereby accurately controlling the test vehicle's impact with the target collision point, ensuring that the test vehicle efficiently completes the collision test task.

[0018] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above and other objects, features and advantages of the present application will become more apparent by describing in more detail exemplary embodiments of the present application in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments of the present application.

[0020] Figure 1 1 is a flow chart of a vehicle collision test method shown in an embodiment of the present application; Figure 2 1 is another flow chart of a vehicle collision test method shown in an embodiment of the present application; Figure 3 is a structural diagram of a collision test system shown in an embodiment of the present application; Figure 4 It is a structural diagram of an electronic device shown in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The following describes embodiments of the present application in more detail with reference to the accompanying drawings. Although the accompanying drawings illustrate embodiments of the present application, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. Rather, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0022] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. As used in this application and the appended claims, the singular forms "a," "an," "the," and "the" are intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0023] It should be understood that although the terms "first", "second", "third", etc. may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0024] With the continuous development of intelligent driving technology, automatic emergency braking systems have become an important component in improving vehicle safety and reducing collision risks. When a potential collision risk is identified, the automatic emergency braking system can proactively implement braking intervention, thereby reducing the probability of a collision.

[0025] In practical applications, vehicles used in AEB testing scenarios are typically equipped with a driving robot or autonomous driving platform to precisely control the steering wheel, accelerator, and brakes to achieve repeatable and controllable collision maneuvers. However, relevant collision testing technologies generally suffer from insufficient positioning accuracy, making it difficult to meet the requirements of AEB performance evaluation.

[0026] In response to the above problems, an embodiment of the present application provides a vehicle collision test method, which can significantly improve the positioning accuracy of the collision test and efficiently complete the collision test task.

[0027] The technical solutions of the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0028] Figure 1 It is a flow chart of a vehicle collision test method shown in an embodiment of the present application.

[0029] See also Figure 1 , applied to a collision test system, the collision test system being in communication with a test vehicle, and comprising at least the following steps: Step 101: Receive a collision test task.

[0030] In the embodiments of this application, the crash test system refers to a system that combines an inertial navigation data processing device, a machine vision enhancement device, and a posture prediction model. Rather than being installed directly inside the test vehicle, the system is deployed on a central server at the AEB testing site. This system utilizes the high-performance servers deployed at the AEB testing site to process data and complete crash testing tasks. For example, industrial computers or desktop computers perform high-load computing tasks such as image processing, path prediction, and control decision-making. The processed results are then transmitted to the test vehicle via wireless communication, enabling precise control of its steering wheel and vehicle speed.

[0031] The test vehicle body only needs to be configured with basic actuators such as servos, motor drivers, and a low-cost single-chip microcomputer control system to receive server instructions and execute actions.

[0032] This deployment method not only significantly reduces the risk of damage to high-value computing equipment during crash tests, but also reduces the complexity and cost of vehicle-mounted systems and improves the flexibility and stability of crash test systems.

[0033] The collision test system can receive a collision test task and, based on the collision test task, generate instructions to control the test vehicle to drive along the expected collision path to the target collision point. During the test vehicle's driving process, the collision test system continuously obtains relevant posture data and continuously corrects the vehicle's current posture.

[0034] In this application, a crash test task refers to a task used to verify whether a test vehicle can accurately impact a set target collision point along an expected collision path under a specific control strategy. It is understood that the task carries the target collision information required by the test vehicle, including but not limited to the target collision point set on the crash test track, the expected collision path, and the expected collision type. For example, the target collision point is the collision object at the end of the crash test track, such as a stationary vehicle, dummy, or other obstacle.

[0035] As an example, the tester sets the target collision point, expected collision path, and expected collision type through the test control platform to initiate a collision test task. After the collision test system receives the collision test task, it learns from the collision test task that the target collision point of this test is a stationary dummy, and the expected collision type requires the test vehicle to accurately collide head-on with the collision object.

[0036] Step 102 : Based on the first visual image data collected by the test vehicle, the corresponding first inertial navigation data is corrected to obtain a first collision posture.

[0037] In an embodiment of the present application, the collision test system is communicatively connected to the test vehicle, so that the collision test system can send a data request to the test vehicle to obtain the first visual image data and first inertial navigation data collected by the test vehicle, and use the first visual image data to correct the corresponding first inertial navigation data to obtain the first collision posture.

[0038] The first visual image data refers to an image sequence and corresponding image data reflecting the vehicle's road environment, which is collected by a high-definition camera or other image sensors while the test vehicle is driving.

[0039] The first inertial navigation data refers to the position data of the test vehicle collected in real time through inertial navigation, including physical quantities such as acceleration, angular velocity, heading angle, attitude angle, speed and displacement.

[0040] The first collision posture refers to posture data obtained after error correction is performed on the first inertial navigation data by combining the first visual image data at a specific moment, including the position and posture angle of the test vehicle at a specific moment and location.

[0041] In actual applications, although the inertial navigation of the test vehicle can continuously record the vehicle's running trajectory and posture changes, it has the problem of cumulative error. Therefore, this application uses the first visual image data to correct the first inertial navigation data, which is beneficial to improving the positioning accuracy of the test vehicle.

[0042] Step 103 : When it is detected that the test vehicle enters the preset critical collision area in the first collision posture, the corresponding second inertial navigation data is corrected based on the second visual image data collected by the test vehicle to obtain a second collision posture.

[0043] In an embodiment of the present application, when the collision test system detects that the test vehicle enters a preset critical collision area in a first collision posture, the corresponding second inertial navigation data can be corrected based on the second visual image data collected by the test vehicle to obtain a second collision posture.

[0044] Among them, the critical collision area refers to the area on the collision test track close to the target collision point. In this area, this application adds dense visual feature points to facilitate the collision test system to identify in advance that the test vehicle has entered the critical collision area and needs to further correct the current posture of the test vehicle.

[0045] The second visual image data refers to the image sequence and corresponding image data collected by the test vehicle's high-definition camera or other image sensor as the test vehicle approaches the target collision point, which is used to further improve the accuracy of visual positioning during the critical collision stage.

[0046] The second inertial navigation data refers to the posture data including acceleration, angular velocity, heading angle, attitude angle, speed and displacement, which are collected in real time by the inertial navigation of the test vehicle while the test vehicle enters the critical collision area. It is used to reflect the posture of the vehicle in the critical stage before the collision.

[0047] The second collision posture refers to the posture of the test vehicle obtained after error correction of the second inertial navigation data based on the second visual image data in the critical collision area, which serves as the basis for subsequent prediction of the target collision posture.

[0048] It is worth noting that the difference between the first visual image data and the second visual image data lies in that the two are collected at different times and locations. The first visual image data is image data collected from the test vehicle in the front area of ​​the collision test track, which is relatively far away from the target collision point, while the second visual image data is image data collected from the test vehicle in the middle and rear areas of the collision test track, which is relatively close to the target collision point. It contains more environmental features and has higher accuracy.

[0049] Correspondingly, the first inertial navigation data corresponding to the first visual image data is navigation data collected by the test vehicle in the front area of ​​the collision test track, and the second inertial navigation data corresponding to the second visual image data is navigation data collected by the test vehicle in the middle and rear areas of the collision test track.

[0050] The difference between the first and second collision poses lies in their differing precision. The first collision pose is relatively inaccurate, providing a rough guide to the test vehicle's intended test path toward the critical collision zone. The second collision pose, obtained by further correcting the navigation data using closer, more detailed second visual image data, offers significantly higher precision. In other words, the first collision pose serves more as a reference for mid- to long-range path navigation, while the second collision pose serves more as a basis for precise dynamic control at closer ranges.

[0051] Step 104 : predicting a target collision posture of the test vehicle based on the second collision posture.

[0052] In an embodiment of the present application, the collision test system can predict the target collision posture of the test vehicle based on the second collision posture.

[0053] The target collision posture refers to the final vehicle position and posture of the test vehicle before it hits the target collision point, which is used to control the test vehicle to hit the target collision point.

[0054] As an example, after obtaining the target collision posture, the collision test system can compare the target collision posture with the current real-time posture, taking into account the current state of the test vehicle. It can then calculate the position error and attitude error between the vehicle's current position and the target posture. Based on the position error and attitude error, the system generates a steering wheel angle command and a speed control command. These commands are packaged and transmitted to the test vehicle via a wireless communication module. The execution unit in the test vehicle receives and executes the steering wheel angle command and speed control command to impact the target collision point.

[0055] In an embodiment of the present application, the technical solution of the present application is applied to a collision test system, which is communicatively connected to a test vehicle and receives a collision test task. The collision test task includes a target collision point set on a collision test track. Based on first visual image data collected by the test vehicle, the corresponding first inertial navigation data is corrected to obtain a first collision posture. When it is detected that the test vehicle enters a preset critical collision area with the first collision posture, the corresponding second inertial navigation data is corrected based on second visual image data collected by the test vehicle to obtain a second collision posture. According to the second collision posture, the target collision posture of the test vehicle is predicted, and the target collision posture is used to control the test vehicle to collide with the target collision point.

[0056] Compared to related technologies, this application achieves multi-stage pose correction during the collision test process by combining visual image data collected by the test vehicle with inertial navigation data, effectively improving positioning accuracy. Furthermore, when the test vehicle enters the critical collision zone, the target collision pose is further predicted based on the corrected collision pose, thereby accurately controlling the test vehicle's impact with the target collision point, ensuring that the test vehicle efficiently completes the collision test task.

[0057] Figure 2 This is another flow chart of a vehicle collision test method shown in an embodiment of the present application. Figure 2 relatively Figure 1 The technical solutions of the embodiments of the present application are described in more detail.

[0058] In order to facilitate understanding of the vehicle collision test method of the embodiment of the present application, the connection structure and connection relationship between the collision test system of the embodiment of the present application and the test vehicle are explained in detail below.

[0059] In this application, the crash test system is deployed on a central server at the AEB test site, which includes at least an inertial navigation data processing device, a machine vision enhancement device, and a posture prediction model. The central server refers to a high-performance server, such as a high-computing computer.

[0060] The test vehicle is equipped with an inertial navigation system and an actuator unit inside, and multiple machine vision image acquisition sensors on the outside. The inertial navigation data collected by the inertial navigation system and the visual image data collected by the machine vision sensors can be uploaded to a central server. The actuator unit includes an actuator and a single-chip microcontroller system. The machine vision acquisition sensors include various types of image acquisition devices, such as high-definition cameras and IP cameras (network cameras).

[0061] Bidirectional data transmission between the test vehicle and the central server is mainly achieved through the WiFi-CAN communication device, where the WiFi-CAN communication device is a combination of a wireless LAN module and a controller LAN module. It can send the control strategy, steering angle, vehicle speed instructions and other contents calculated by the server to the execution unit on the vehicle through WiFi-CAN messages.

[0062] Furthermore, the inertial navigation data processing device is capable of receiving and preprocessing the inertial navigation data collected by the test vehicle using the inertial navigation equipment, and the inertial navigation data processing device is connected to the machine vision enhancement device, and is capable of sending the inertial navigation data to the machine vision enhancement device, and receiving the error value of the inertial navigation data corrected by the machine vision enhancement device in combination with the visual image data.

[0063] In addition to receiving visual image data from the image acquisition sensor, the machine vision enhancement device can also receive navigation data from the inertial navigation data processing device, correct the inertial navigation data through the internally configured error correction model, and output the collision posture.

[0064] The pose prediction model predicts the vehicle's current position and pose more accurately based on visual image data and inertial navigation data.

[0065] As an example, the present application can further refine the internal structure of the collision test system according to the functions of the above-mentioned devices and models. The collision test system can be composed of a collision test task receiving module, a first pose correction module, a second pose correction module, and a target collision pose prediction module.

[0066] Among them, the first posture correction module and the second posture correction module can be jointly implemented by the inertial navigation data processing device and the machine vision enhancement device, the target collision posture prediction module can be implemented by calling the posture prediction model deployed, and the collision test task receiving module can be implemented by the central server with its built-in communication and task management functions.

[0067] Next, the collision test process of the present application is further described in conjunction with the aforementioned collision test system and test vehicle, which at least includes the following steps.

[0068] Step 201: Receive a collision test task.

[0069] In an embodiment of the present application, a collision test system may receive a collision test task, where the collision test task carries a target collision point and a plurality of visual feature points located on a collision test track.

[0070] Among them, visual feature points refer to feature points identified by machine vision enhancement devices, including runway markings and target collision points.

[0071] Runway markings include at least lines, areas, lane markings, arrows and other visual guidance elements. For example, the red line at the end indicates that deceleration begins 5m from the collision point, and the side lines indicate lane boundaries.

[0072] The target collision point is a special visual mark indicating the collision position. For example, a special bull's-eye shape is affixed to the target collision point position to facilitate visual recognition and distance estimation.

[0073] In this application, different numbers and types of visual feature points are set in different areas of the collision test track to meet the differentiated requirements of the test vehicle for positioning accuracy at different stages.

[0074] As an example, a moderate density of runway markings are set up in the front area of ​​the collision test track to guide the test vehicle along the expected collision path and reduce the error range.

[0075] In the critical collision area close to the target collision point, visual feature points with higher density and stronger recognition are added to ensure high-precision posture calibration of the vehicle's inertial navigation posture at the critical stage before the collision.

[0076] It is worth noting that the position and posture of the target collision point and each visual feature point are pre-calibrated and known, that is, their position information and posture information in the test coordinate system are determined and can be directly used for subsequent calculations and comparisons.

[0077] Step 202: Acquire first visual image data and first inertial navigation data.

[0078] In an embodiment of the present application, when the collision test system detects that the test vehicle is traveling along the expected collision path but has not yet reached the critical collision area, the first visual image data and first inertial navigation data before the critical collision area can be obtained first.

[0079] As an example, the starting position of the robot-driven vehicle is approximately 50 meters away from the target collision point, such as a soft vehicle or dummy. When the collision test system receives the collision test task, it can first use inertial navigation technology for preliminary path guidance. This process does not require the participation of visual recognition technology and can control the error between the actual driving path and the expected collision path to within approximately 10 cm. Next, the first visual image data collected by the test vehicle using visual recognition technology and the first inertial navigation data collected using inertial navigation technology are obtained. Finally, the inertial navigation data processing device and machine vision enhancement device are used to perform data preprocessing such as denoising and feature enhancement.

[0080] Step 203 : Correcting the first inertial navigation data based on the first visual image data including the first visual feature point to obtain a first collision posture.

[0081] In actual use, the test vehicle will continuously upload the first visual image data and the first inertial navigation data. The collision test system needs to select a suitable visual image frame from several frames. It can be understood that the collision test system can perform feature recognition on the visual image frame. If the first visual feature point is recognized, the corresponding visual image frame is selected, and the associated first visual image data is used to correct the first inertial navigation data to obtain the first collision posture.

[0082] In an optional example of the present application, based on the first visual image data collected by the test vehicle, the corresponding first inertial navigation data is corrected, and the process of obtaining the first collision posture at least includes: obtaining the first visual image data and the first inertial navigation data collected when the test vehicle is traveling along the collision test track, performing feature extraction on the first visual image data, obtaining the first visual feature point and the first visual feature data of the first visual feature point, and using the first visual feature data to correct the corresponding first inertial navigation data to obtain the first collision posture.

[0083] Furthermore, the first visual feature data is used to correct the corresponding first inertial navigation data to obtain a first collision posture, including: extracting the first acquisition time point and the first feature point deviation posture of the first visual feature point from the first visual feature data, extracting the first measurement posture corresponding to the first acquisition time point from the first inertial navigation data, and correcting the first measurement posture using the first feature point deviation posture to obtain the first collision posture.

[0084] The first acquisition time point refers to the timestamp when the test vehicle acquires the first visual image containing the first visual feature point during driving. The timestamp is used to establish a time synchronization relationship between the visual image and the inertial navigation data.

[0085] The first feature point deviation pose refers to the spatial difference between the first visual feature point identified in the visual image and its known reference pose, including position deviation and pose deviation.

[0086] The first measured posture refers to the vehicle posture outputted by the inertial navigation device inside the test vehicle at the first acquisition time point.

[0087] Taking the first visual feature points as runway markings as an example, the collision test system acquires several frames of first visual images, performs low-level feature extraction through image processing, and then uses an error correction model for high-level recognition and positioning. Deep learning algorithms such as YOLO (You Only Look Once) and Faster R-CNN (Faster Region-based Convolutional Neural Network) are used to perform real-time recognition and positioning of runway markings in visual images. These visual landmarks can be irregular shapes or natural features, and they are associated with inertial navigation data.

[0088] Furthermore, the actual pose (i.e., spatial position and orientation) of the runway marker in the current visual image frame is calculated and compared with a reference pose of the runway marker pre-calibrated on the collision test track to obtain the deviation pose of the first feature point of the runway marker. For example, this can include the offset between the spatial position of the runway marker in the current visual image frame and the reference position, the deviation of the runway marker's orientation angle from the reference orientation, and the difference between the depth distance inferred based on the camera's viewing angle and the known marker size and the actual set reference distance.

[0089] During the correction process, in order to ensure the effectiveness of the correction, the acquisition time point is extracted from the visual image. At the same acquisition time point, the first feature point deviation posture is used to correct the first measured posture to obtain the first collision posture.

[0090] In this application, correction methods include linear superposition correction and fusion correction.

[0091] Linear superposition correction: Directly add the first feature point deviation pose to the first measured pose. For example, if the first visual feature point deviation pose is a vehicle position offset of 3 cm to the right and a yaw angle deviation of +1.5°, then directly add these two error values ​​to the original inertial navigation position and angle respectively to obtain the first collision pose.

[0092] Fusion correction: The first feature point deviation pose and the first measured pose are used as input data and input into a pre-set fusion algorithm to obtain an accurate first collision pose.

[0093] It can be seen that the correction process of this application is essentially a vision + inertial navigation fusion correction, which uses the visual image data collected by the machine vision acquisition sensor as the correction basis to correct the measurement results of inertial navigation, thereby achieving both stability and accuracy.

[0094] Step 204 : Detecting that the test vehicle enters the critical collision area, and acquiring second visual image data and second inertial navigation data.

[0095] As can be seen from the above description of the key collision area, in order to improve the correction accuracy, this application adds more and more recognizable second visual feature points in the key collision area before testing.

[0096] When it is detected that the test vehicle enters the critical collision area, which means a time period close to the collision point, the collision test system can obtain second visual image data and second inertial navigation data.

[0097] Step 205 : Correct the second inertial navigation data based on the second visual image data including the second visual feature point to obtain a second collision posture.

[0098] In an optional example of the present application, when it is detected that the test vehicle enters a preset critical collision area in a first collision posture, the corresponding second inertial navigation data is corrected based on the second visual image data collected by the test vehicle, and the process of obtaining the second collision posture at least includes: obtaining the second visual image data of the test vehicle traveling in the first collision posture, if a second visual feature point is identified from the second visual image data, and the collision distance between the test vehicle and the target collision point is less than a set distance threshold, it is determined that the test vehicle enters the critical collision area, and the second visual feature data corresponding to the second visual feature point is obtained, and the corresponding second inertial navigation data is corrected using the second visual feature data to obtain the second collision posture.

[0099] Furthermore, the second visual feature data is used to correct the corresponding second inertial navigation data to obtain a second collision posture, including: extracting the second acquisition time point and the second feature point deviation posture of the second visual feature point from the second visual feature data, extracting the second measurement posture corresponding to the second acquisition time point from the second inertial navigation data, and using the second feature point deviation posture to correct the second measurement posture to obtain the second collision posture.

[0100] The second acquisition time point refers to the timestamp when the test vehicle enters the critical collision area and acquires the second visual image containing the second visual feature point. The timestamp is used to establish a time synchronization relationship between the visual image and the inertial navigation data.

[0101] The second feature point deviation pose refers to the spatial difference between the second visual feature point identified in the visual image and its known reference pose, including position deviation and pose deviation.

[0102] The second measured posture refers to the vehicle posture outputted by the inertial navigation device inside the test vehicle at the second acquisition time point.

[0103] As an example, if at least one second visual feature point is successfully identified from the collected second visual image data, and the calculated collision distance between the test vehicle and the target collision point is less than the set distance threshold, the test vehicle is determined to have entered the critical collision area, thereby triggering a more accurate correction process and improving the accuracy of subsequent collision control.

[0104] It should be understood that the principle of the above-mentioned process of correcting the second inertial navigation data based on the second visual feature data to obtain the second collision posture is similar to the process of correcting the first inertial navigation data based on the first visual feature data, and also includes the steps of feature extraction, time point alignment, posture deviation calculation and correction. The relevant technical details can be found in the above content and will not be repeated here.

[0105] In this application, first, the first stage of posture correction is completed based on the first visual image data and the first inertial navigation data to obtain a preliminary and accurate first collision posture. Subsequently, as the test vehicle approaches the collision point, the second stage of posture correction is performed by identifying the more densely distributed second visual feature points to further obtain a higher-precision second collision posture. Since both correction processes fully integrate high-confidence visual feature data and inertial navigation data, the collision test system can effectively correct posture deviations caused by inertial accumulation errors, image drift errors, and scene interference. At this point, the posture error of the test vehicle is effectively suppressed to within 5 cm, greatly improving the accuracy and reliability of collision control.

[0106] Step 206 : Control the test vehicle to automatically drive according to the second collision posture to obtain target visual image data and target inertial navigation data.

[0107] When the test vehicle enters the critical collision area in the second collision posture, it can be considered that the vehicle's current position and orientation are relatively close to the expected final collision path and collision point. Therefore, the target visual image data and target inertial navigation data corresponding to the test vehicle's automatic driving in the second collision posture can be used as the last correction data.

[0108] Step 207 : Input the target visual image data and the target inertial navigation data into a preset posture prediction model to predict the target collision posture.

[0109] In this application, the target visual image data and target inertial navigation data are used as model input data and input into a preset posture prediction model to predict the target collision posture.

[0110] In an optional example of the present application, the process of the posture prediction model predicting the target collision posture of the test vehicle based on the second collision posture includes at least: obtaining target visual image data and target inertial navigation data of the test vehicle traveling in the second collision posture, performing data preprocessing on the target visual image data and target inertial navigation data respectively, generating model input data, inputting the model input data into a preset posture prediction model, and outputting the target collision posture.

[0111] Among them, the data preprocessing method includes at least image denoising, image enhancement, interpolation and padding, and filtering noise reduction. The target visual image data and target inertial navigation data after data preprocessing are aligned to a unified timestamp to generate model input data.

[0112] Among them, the posture prediction model is constructed as follows: S1, build the initial neural network model and training sample set.

[0113] Optionally, the initial neural network model can be a convolutional neural network (CNN), which extracts feature points through multi-layer convolution, pooling, and other operations to effectively identify the test vehicle's environment. Alternatively, the initial neural network model can be a recurrent neural network (RNN), leveraging its advantages in processing time series data to dynamically model inertial navigation data, thereby more accurately capturing changes in the test vehicle's position and posture.

[0114] The training sample set includes at least historical inertial navigation data, historical visual image data and historical real poses.

[0115] Among them, the historical inertial navigation data records the recorded acceleration, angular velocity, heading angle, position change and other data.

[0116] Historical visual image data records visual scene information at different time points and in different areas, and includes visual feature points such as runway markings, arrows, sidelines, and targets.

[0117] The historical real pose is the actual position and pose information of the test vehicle along the historical path, which is usually obtained through a high-precision positioning system such as GPS (Global Positioning System) + RTK (Real-Time Kinematic) or an optical positioning system.

[0118] S2, uses the training sample set to train the initial neural network model to obtain the predicted pose.

[0119] The initial neural network model is supervised and trained using the training sample set. The loss function is used to measure the difference between the predicted pose and the actual pose, and the model parameters are continuously optimized through back propagation.

[0120] S3, calculate the training error between the predicted pose and the historical true pose. When the training error converges to the preset threshold, the model training is completed and the pose prediction model is generated.

[0121] During the training process, the error between the predicted output and the true value is continuously calculated. When the training error converges to the preset threshold, the training ends and the current model is determined as the pose prediction model.

[0122] As an example, during the model training phase, a large amount of labeled training data is prepared. This training data includes historical inertial navigation data such as acceleration and angular velocity output by the inertial navigation system, as well as image frames or image sequences captured by the machine vision sensor system. All historical data is unfused. At the same time, each frame of image or each point in time of inertial navigation data corresponds to high-precision reference pose information. This high-precision reference pose information is obtained through means such as high-precision GPS+RTK or optical positioning systems and serves as the true label for supervised learning.

[0123] The core of data labeling is to establish a mapping relationship between "raw input data" and "real-world position and pose" to form training sample pairs. After inputting this labeled data into the deep learning model, the network parameters are continuously adjusted through the backpropagation algorithm. The model gradually learns how to accurately estimate the vehicle's real-time position and pose from the raw inertial navigation and vision data, significantly reducing prediction errors and improving the overall accuracy of navigation and control.

[0124] In addition, to further improve model performance, a data feeding mechanism is introduced to continuously input a large amount of accurately labeled raw data into the model during training. The specific process includes: First, the inertial navigation data and machine vision image data are used as input for forward propagation to generate prediction results of the vehicle position and posture, and the error between the prediction results and the true label is calculated. Common loss functions include mean square error, attitude angle error or quaternion loss.

[0125] Then, the error gradient is calculated through the back-propagation algorithm to evaluate the contribution of each layer parameter to the error. The network parameters are updated using optimization algorithms, such as Adam (adaptive moment estimation algorithm) and stochastic gradient descent (SGD), to adjust the prediction direction and amplitude.

[0126] After multiple rounds of iterative training, the model parameters gradually converged, obtaining a more stable and reliable pose prediction capability, enabling it to accurately output high-precision vehicle pose estimates when faced with new input data.

[0127] This application optimizes the fusion results between inertial navigation data and visual image data, controls the collision point positioning error within 2cm, fully leverages the advantages of large-scale data-driven and modern computing capabilities, and greatly improves the accuracy and reliability of the AEB collision test driving robot's steering wheel and vehicle speed control.

[0128] Step 208 : Generate a control instruction corresponding to the target collision posture to control the test vehicle to collide with the target collision point.

[0129] In the embodiment of the present application, the collision test system generates precise control instructions based on the target collision posture predicted in step 207 and the current motion state of the test vehicle to drive the vehicle to achieve precise collision.

[0130] In this application, the control instructions generated by the collision test system are mainly steering wheel control instructions and vehicle speed control instructions.

[0131] As an example, assuming that the current heading angle of the test vehicle is predicted to be 5°, and the expected collision path requires the test vehicle to go straight ahead, that is, the heading angle is 0°, then the heading angle deviation is converted into a steering wheel angle command, and the steering wheel angle command is sent to the vehicle execution unit to control the steering wheel adjustment of the test vehicle to ensure that the vehicle travels towards the predetermined collision path.

[0132] Similarly, assuming that the current speed of the test vehicle is predicted to be 20 km / h and the target is 10 km / h, it is calculated that a deceleration of 10 km / h is required. The speed deviation is mapped to a braking force command, and the braking force command is sent to the vehicle execution unit to control the vehicle to decelerate at the expected speed.

[0133] In an embodiment of the present application, by combining an inertial navigation data processing device, a machine vision enhancement device, and a posture prediction model, fusion analysis and collision posture prediction are achieved to generate accurate control instructions and strategies.

[0134] In addition, the collision test system is deployed in the AEB test site rather than inside the test vehicle, avoiding the direct deployment of expensive, bulky and fragile computing equipment inside the test vehicle, reducing the risk of equipment damage and maintenance costs, and improving the flexibility and scalability of the system.

[0135] Corresponding to the aforementioned application function implementation method embodiment, the present application also provides a collision test system, electronic equipment and corresponding embodiments.

[0136] Figure 3 It is a structural diagram of a collision test system shown in an embodiment of the present application.

[0137] See also Figure 3 The collision test system is in communication with the test vehicle; wherein the system includes: A collision test task receiving module 301 is used to receive a collision test task, where the collision test task includes a target collision point set on a collision test track; A first posture correction module 302 is configured to correct the corresponding first inertial navigation data based on the first visual image data collected by the test vehicle to obtain a first collision posture; A second posture correction module 303 is configured to correct the corresponding second inertial navigation data based on the second visual image data collected by the test vehicle to obtain a second collision posture when the test vehicle is detected to have entered a preset critical collision area in the first collision posture; The target collision posture prediction module 304 is used to predict the target collision posture of the test vehicle according to the second collision posture. The target collision posture is used to control the test vehicle to collide with the target collision point.

[0138] In an optional example of the present application, the first pose correction module 302 includes: a first data acquisition submodule, configured to acquire first visual image data and first inertial navigation data collected when the test vehicle travels along the collision test track; A first feature extraction submodule is used to extract features from the first visual image data to obtain first visual feature points and first visual feature data of the first visual feature points; The first posture correction submodule is used to correct the corresponding first inertial navigation data using the first visual feature data to obtain a first collision posture.

[0139] In an optional example of the present application, the first pose correction submodule is used to: Extracting a first acquisition time point and a first feature point deviation posture of a first visual feature point from the first visual feature data; Extracting a first measurement pose corresponding to a first acquisition time point from the first inertial navigation data; The first measured posture is corrected using the first feature point deviation posture to obtain the first collision posture.

[0140] In an optional example of the present application, the second posture correction module 303 includes: A second data acquisition submodule is used to acquire second visual image data of the test vehicle traveling in the first collision posture; a critical collision area determination submodule, configured to determine that the test vehicle has entered the critical collision area if a second visual feature point is identified from the second visual image data and the collision distance between the test vehicle and the target collision point is less than a set distance threshold; A third data acquisition submodule is used to acquire second visual feature data corresponding to the second visual feature point; The second posture correction submodule is used to correct the corresponding second inertial navigation data using the second visual feature data to obtain a second collision posture.

[0141] In an optional example of the present application, the second posture correction submodule is used to: Extracting a second acquisition time point and a second feature point deviation posture of the second visual feature point from the second visual feature data; Extracting a second measurement pose corresponding to a second acquisition time point from the second inertial navigation data; The second measured posture is corrected using the second feature point deviation posture to obtain the second collision posture.

[0142] In an optional example of the present application, the target collision posture prediction module 304 includes: a fourth data acquisition submodule, configured to acquire target visual image data and target inertial navigation data of the test vehicle traveling in a second collision posture; The input data generation submodule is used to perform data preprocessing on the target visual image data and the target inertial navigation data respectively to generate model input data; The target collision posture prediction submodule is used to input model input data into a preset posture prediction model and output the target collision posture.

[0143] In an optional example of this application, the pose prediction model is constructed as follows: Constructing an initial neural network model and a training sample set; wherein the training sample set includes at least historical inertial navigation data, historical visual image data, and historical real poses; Use the training sample set to train the initial neural network model to obtain the predicted pose; The training error between the predicted pose and the historical true pose is calculated. When the training error converges to the preset threshold, the model training is completed and the pose prediction model is generated.

[0144] Regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated again here.

[0145] Figure 4 It is a structural diagram of an electronic device shown in an embodiment of the present application.

[0146] See also Figure 4 , the electronic device 400 includes a memory 410 and a processor 420.

[0147] The processor 420 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. Memory 410 may include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage. ROM may store static data or instructions required by processor 420 or other computer modules. Permanent storage may be a readable and writable storage device. Permanent storage may be a non-volatile storage device that retains stored instructions and data even when the computer is powered off. In some embodiments, the permanent storage device utilizes a mass storage device (e.g., a magnetic or optical disk, flash memory). In other embodiments, the permanent storage device may be a removable storage device (e.g., a floppy disk, optical drive). System memory may be a readable and writable storage device or a volatile readable and writable storage device, such as dynamic random access memory (DRAM). System memory may store some or all instructions and data required by the processor during operation. Furthermore, memory 410 may include any combination of computer-readable storage media, including various types of semiconductor memory chips (e.g., DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), as well as magnetic disks and / or optical disks. In some embodiments, the memory 410 may include a readable and / or writable removable storage device, such as a compact disc (CD), a read-only digital versatile disc (e.g., DVD-ROM, dual-layer DVD-ROM), a read-only Blu-ray disc, an ultra-density optical disc, a flash memory card (e.g., SD card, mini SD card, Micro-SD card, etc.), a magnetic floppy disk, etc. Computer-readable storage media do not contain carrier waves and transient electronic signals transmitted wirelessly or wired.

[0148] The memory 410 stores executable codes. When the executable codes are processed by the processor 420 , the processor 420 may execute part or all of the above-mentioned methods.

[0149] In addition, the method according to the present application may also be implemented as a computer program or a computer program product, which includes computer program code instructions for executing some or all of the steps in the above method of the present application.

[0150] Alternatively, the present application can also be implemented as a computer-readable storage medium (or non-transitory machine-readable storage medium or machine-readable storage medium), which stores executable code (or computer program or computer instruction code) and, when executed by a processor of an electronic device (or server, etc.), enables the processor to perform part or all of the steps of the above-mentioned method according to the present application.

[0151] The present application also provides a computer program product, which includes computer instructions, and when the computer instructions are executed by a processor, the method described above is implemented.

[0152] The embodiments of the present application have been described above. The above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to the technology in the market, or to enable other persons skilled in the art to understand the embodiments disclosed herein.

Claims

1. A vehicle collision test method, characterized in that: Applied to a collision test system, the collision test system being communicatively connected to a test vehicle, the method comprising: receiving a collision test task, wherein the collision test task includes a target collision point set on a collision test track; Based on the first visual image data collected by the test vehicle, correcting the corresponding first inertial navigation data to obtain a first collision posture; When detecting that the test vehicle enters a preset critical collision area in the first collision posture, correcting corresponding second inertial navigation data based on second visual image data collected by the test vehicle to obtain a second collision posture; A target collision posture of the test vehicle is predicted based on the second collision posture, and the target collision posture is used to control the test vehicle to collide with the target collision point.

2. The method according to claim 1, characterized in that The method of correcting corresponding first inertial navigation data based on first visual image data collected by the test vehicle to obtain a first collision posture includes: Acquiring first visual image data and first inertial navigation data collected when the test vehicle travels along the collision test track; Performing feature extraction on the first visual image data to obtain a first visual feature point and first visual feature data of the first visual feature point; The first visual feature data is used to correct the corresponding first inertial navigation data to obtain the first collision posture.

3. The method according to claim 2, characterized in that The method of correcting the corresponding first inertial navigation data using the first visual feature data to obtain the first collision posture includes: Extracting a first acquisition time point and a first feature point deviation posture of the first visual feature point from the first visual feature data; Extracting a first measurement pose corresponding to the first acquisition time point from the first inertial navigation data; The first measured posture is corrected using the first feature point deviation posture to obtain the first collision posture.

4. The method according to claim 1, wherein When detecting that the test vehicle enters a preset critical collision area in the first collision posture, correcting corresponding second inertial navigation data based on second visual image data collected by the test vehicle to obtain a second collision posture includes: Acquiring second visual image data of the test vehicle traveling in the first collision posture; If a second visual feature point is identified from the second visual image data and the collision distance between the test vehicle and the target collision point is less than a set distance threshold, determining that the test vehicle has entered the critical collision area; Acquire second visual feature data corresponding to the second visual feature point; The second visual feature data is used to correct the corresponding second inertial navigation data to obtain the second collision posture.

5. The method according to claim 4, characterized in that The method of correcting the corresponding second inertial navigation data using the second visual feature data to obtain the second collision posture includes: Extracting a second acquisition time point and a second feature point deviation posture of the second visual feature point from the second visual feature data; Extracting a second measurement pose corresponding to the second acquisition time point from the second inertial navigation data; The second measured posture is corrected using the second feature point deviation posture to obtain the second collision posture.

6. The method according to claim 1, characterized in that Predicting a target collision posture of the test vehicle according to the second collision posture includes: Acquiring target visual image data and target inertial navigation data of the test vehicle traveling in the second collision posture; performing data preprocessing on the target visual image data and the target inertial navigation data respectively to generate model input data; The model input data is input into a preset posture prediction model to output the target collision posture.

7. The method according to claim 6, characterized in that The posture prediction model is constructed as follows: Constructing an initial neural network model and a training sample set; wherein the training sample set includes at least historical inertial navigation data, historical visual image data, and historical real poses; Using the training sample set to train the initial neural network model to obtain a predicted pose; The training error between the predicted posture and the historical true posture is calculated. When the training error converges to a preset threshold, the model training is completed and the posture prediction model is generated.

8. A collision test system, characterized in that: The collision test system is in communication with the test vehicle; wherein the system comprises: A collision test task receiving module, configured to receive a collision test task, wherein the collision test task includes a target collision point located on a collision test track; a first posture correction module, configured to correct corresponding first inertial navigation data based on first visual image data collected by the test vehicle to obtain a first collision posture; a second posture correction module, configured to correct corresponding second inertial navigation data based on second visual image data collected by the test vehicle to obtain a second collision posture when detecting that the test vehicle enters a preset critical collision area in the first collision posture; A target collision posture prediction module is used to predict a target collision posture of the test vehicle according to the second collision posture, and the target collision posture is used to control the test vehicle to collide with the target collision point.

9. An electronic device, characterized in that: include: processor; as well as A memory having executable codes stored thereon, which, when executed by the processor, causes the processor to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having executable codes stored thereon, wherein when the executable codes are executed by a processor of an electronic device, the processor is caused to execute the method according to any one of claims 1 to 7.

Images