Turning collision detection method, apparatus, device, and product

Abstract

The application discloses a turning collision detection method, device, equipment and product, relates to the technical field of vehicle safety, and the method is applied to a vehicle and comprises the following steps: when detecting that the vehicle is turning, acquiring vehicle driving state data, three-dimensional point cloud data and visual image data through a sensor module; acquiring fusion target data by performing space-time alignment and fusion on the three-dimensional point cloud data and the visual image data through a sensor fusion module; generating a three-dimensional stereoscopic attention region according to the vehicle driving state data, filtering the fusion target data through the three-dimensional stereoscopic attention region, and obtaining potential collision target data; performing space interference analysis on the potential collision target data, obtaining a collision point coordinate and a collision point speed, generating an operation decision through the collision point coordinate and the collision point speed, and executing the operation decision. The method solves the problem that three-dimensional perception is insufficient in the inner wheel difference area of the turning vehicle, and thus the vehicle warning and braking are inaccurate.

Description

Technical Field

[0001] This application relates to the field of vehicle safety technology, and in particular to a steering collision detection method, device, equipment, and product. Background Technology

[0002] Large vehicles such as trucks, buses, and cement mixers, due to their large size and wide blind spots, create a significant inner wheel difference area during turns. This area poses a serious safety threat to vulnerable road users such as pedestrians and non-motorized vehicles, and is prone to causing traffic accidents. Currently, the industry widely uses blind spot monitoring and turning assistance systems to reduce this risk. These systems are mainly based on millimeter-wave radar or visual camera technology and are the mainstream technical solution for the safety protection of large vehicles when turning. They are used to monitor potential obstacles during turns and issue warnings to help drivers avoid collision risks.

[0003] However, existing systems based on millimeter-wave radar or visual cameras have significant limitations in terms of technical implementation. First, the existing perception dimensions are limited, lacking three-dimensional information of the target, making it difficult to obtain the true three-dimensional contour and size of obstacles, which easily leads to false alarms and missed alarms. Second, existing risk models use preset static monitoring areas and cannot dynamically adjust the risk range according to the vehicle's real-time steering angle and driving speed, resulting in deviations from the actual inner wheel difference trajectory. Finally, the alarm threshold is a fixed value, which cannot be adapted to different traffic scenarios such as intersections and schools, easily leading to false alarms or response delays. These limitations result in poor system protection and an inability to effectively avoid the safety risks in the inner wheel difference area.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main purpose of this application is to provide a steering collision detection method, device, equipment and product, which aims to solve the technical problem of inaccurate vehicle warning and braking caused by insufficient three-dimensional perception of the inner wheel difference area when the vehicle is turning.

[0006] To achieve the above objectives, this application proposes a steering collision detection method, which is applied to a vehicle, and includes: When the vehicle is detected to be turning, the sensor module acquires vehicle driving status data, 3D point cloud data, and visual image data. The sensor fusion module performs spatiotemporal alignment and fusion of the 3D point cloud data and visual image data to obtain fused target data. A three-dimensional region of interest is generated based on the vehicle driving status data, and the fused target data is filtered through the three-dimensional region of interest to obtain potential collision target data. Spatial interferometry analysis is performed on the potential collision target data to obtain the collision point coordinates and collision point velocity. Operation decisions are generated and executed based on the collision point coordinates and collision point velocity.

[0007] In one embodiment, the sensor module includes a 3D LiDAR, a fisheye camera, and a vehicle attitude sensor. The step of acquiring vehicle driving status data, 3D point cloud data, and visual image data through the sensor module includes: The three-dimensional lidar is used to perform three-dimensional scanning and point cloud analysis of the vehicle's surrounding environment to obtain three-dimensional point cloud data. The fisheye camera is used to acquire and preprocess images of the vehicle's steering side environment to obtain visual image data. The vehicle speed, steering wheel angle, and wheelbase are collected by the vehicle attitude sensor, and the vehicle driving status data are obtained by integrating the vehicle speed, steering wheel angle, and wheelbase.

[0008] In one embodiment, the step of performing spatiotemporal alignment and fusion of the 3D point cloud data and visual image data using a sensor fusion module to obtain fused target data includes: The three-dimensional point cloud data and visual image data are time-stamped and synchronized to obtain a time synchronization result. Based on the vehicle's coordinate system, the three-dimensional point cloud data and visual image data are transformed into spatial coordinates to obtain spatial synchronization results; Based on the spatial synchronization results and temporal synchronization results, feature fusion is performed on the 3D point cloud data and visual image data to obtain fused target data.

[0009] In one embodiment, the step of generating a three-dimensional region of interest based on the vehicle driving state data includes: Based on the vehicle speed and steering wheel angle in the vehicle driving status data, the turning radius and inner wheel difference trajectory of the vehicle are calculated using the Ackermann steering dynamics model to obtain the inner wheel difference trajectory information; Based on the vehicle's wheelbase, track width, and body height parameters, the inner wheel difference trajectory is extended from a two-dimensional plane to a three-dimensional space to obtain spatial trajectory information; A three-dimensional region of interest is constructed based on the spatial trajectory information.

[0010] In one embodiment, the step of filtering the fused target data through the three-dimensional region of interest to obtain potential collision target data includes: The target's three-dimensional coordinates in the fused target data are spatially matched with the three-dimensional region of interest to obtain the matching result; Based on the matching results, targets within the three-dimensional area of ​​interest are filtered to obtain candidate target data; Invalid targets in the three-dimensional region of interest are removed based on the candidate target data, and the candidate target data is optimized to obtain potential collision target data.

[0011] In one embodiment, the step of performing spatial interferometry analysis on the potential collision target data to obtain the collision point coordinates and collision point velocity includes: The relative distance and relative motion data between the vehicle and the potential target are calculated based on the potential collision target data to obtain motion parameter information; Spatial interference calculations are performed on the three-dimensional contour of the potential target through the three-dimensional region of interest to obtain the collision point coordinates; The velocity at the collision point is calculated using the motion parameters and the coordinates of the collision point.

[0012] In one embodiment, the step of generating and executing an operational decision based on the collision point coordinates and collision point velocity includes: The traffic scene type is obtained by performing AI scene recognition on the visual image data. The initial collision time warning threshold and the initial collision time braking threshold are adjusted according to the traffic scenario type to obtain the final collision time warning threshold and the final collision time braking threshold. The collision time is calculated using the collision point coordinates and the collision point velocity. The collision time is analyzed for decision-making based on the final collision time warning threshold and the final collision time braking threshold. If the collision time does not exceed the final collision time braking threshold, the braking control signal is determined as a decision command, and the decision command is executed. If the collision time does not exceed the final collision time warning threshold but exceeds the final collision time braking threshold, a warning will be sent as a decision instruction and the decision instruction will be executed.

[0013] Furthermore, to achieve the above objectives, this application also proposes a steering collision detection device, which is applied to a vehicle and includes: The acquisition module is used to acquire vehicle driving status data, 3D point cloud data and visual image data through the sensor module when the vehicle is detected to be turning. The fusion module is used to perform spatiotemporal alignment and fusion of the 3D point cloud data and visual image data through the sensor fusion module to obtain fused target data. The filtering module is used to generate a three-dimensional region of interest based on the vehicle driving status data, and to filter the fused target data through the three-dimensional region of interest to obtain potential collision target data. The execution module is used to perform spatial interferometry analysis on the potential collision target data to obtain the collision point coordinates and collision point velocity, and to generate and execute operation decisions based on the collision point coordinates and collision point velocity.

[0014] In addition, to achieve the above objectives, this application also proposes a steering collision detection device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the steering collision detection method as described above.

[0015] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the steering collision detection method described above.

[0016] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the steering collision detection method described above.

[0017] One or more technical solutions proposed in this application have at least the following technical effects: This application proposes a steering collision detection method, apparatus, device, and product. The method is applied to a vehicle. When detecting a vehicle turning, a sensor module acquires vehicle driving state data, 3D point cloud data, and visual image data. A sensor fusion module performs spatiotemporal alignment and fusion of the 3D point cloud data and visual image data to obtain fused target data. A 3D region of interest is generated based on the vehicle driving state data, and the fused target data is filtered through this region to obtain potential collision target data. Spatial interference analysis is performed on the potential collision target data to obtain collision point coordinates and velocity. An operational decision is generated and executed based on the collision point coordinates and velocity. Thus, by acquiring 3D point cloud data and visual image data through a sensor module, 3D environmental perception is achieved, compensating for the lack of 2D perception. Precise 3D target data is then obtained through spatiotemporal fusion. A 3D region of interest is generated based on the driving state, matching the actual inner wheel difference space. Further spatial interference analysis yields the precise collision point and velocity, enabling decision-making for warning or braking. This solves the problem of inaccurate vehicle warnings and braking due to insufficient 3D perception of the inner wheel difference area during vehicle turning, improving the accuracy of steering collision detection. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating an embodiment of the steering collision detection method of this application. Figure 2 This is a schematic diagram illustrating the blind spot of the vehicle involved in the steering collision detection method of this application; Figure 3 This is a schematic diagram of a simplified system architecture involved in the steering collision detection method of this application; Figure 4 This is a schematic diagram illustrating the construction of the three-dimensional region of interest involved in the steering collision detection method of this application; Figure 5 This is a flowchart illustrating Embodiment 2 of the steering collision detection method of this application; Figure 6 This is a schematic diagram illustrating the AI ​​model construction involved in the steering collision detection method of this application; Figure 7 This is a schematic diagram illustrating the AI ​​model detection involved in the steering collision detection method of this application; Figure 8 This is a schematic diagram illustrating the scene recognition and threshold determination involved in the steering collision detection method of this application; Figure 9 This is a schematic diagram illustrating the decision analysis involved in the steering collision detection method of this application; Figure 10 This is a simplified flowchart of the steering collision detection method provided in Embodiment 2 of this application; Figure 11 This is a schematic diagram of the module structure of the steering collision detection device according to an embodiment of this application; Figure 12 This is a schematic diagram of the device structure of the hardware operating environment involved in the steering collision detection method in this application embodiment.

[0021] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0023] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0024] The main solution of this application embodiment is as follows: The three-dimensional LiDAR performs three-dimensional scanning and point cloud analysis of the vehicle's surrounding environment to obtain three-dimensional point cloud data; the fisheye camera performs image acquisition and preprocessing of the vehicle's steering side environment to obtain visual image data; the vehicle's speed, steering wheel angle, and wheelbase are acquired by a vehicle attitude sensor, and the speed, steering wheel angle, and wheelbase are integrated to obtain vehicle driving state data. The three-dimensional point cloud data and visual image data are time-stamped and synchronized to obtain a time synchronization result; spatial coordinate transformation is performed on the three-dimensional point cloud data and visual image data according to the vehicle's coordinate system to obtain a spatial synchronization result; feature fusion is performed on the three-dimensional point cloud data and visual image data based on the spatial synchronization result and the time synchronization result to obtain fused target data. Based on the vehicle speed and steering wheel angle in the vehicle driving state data, the vehicle turning radius and inner wheel difference trajectory are calculated using the Ackermann steering dynamics model to obtain inner wheel difference trajectory information. Based on the vehicle's wheelbase, track width, and vehicle height parameters, the inner wheel difference trajectory is extended from a two-dimensional plane to three-dimensional space to obtain spatial trajectory information. A three-dimensional region of interest is constructed based on the spatial trajectory information. The three-dimensional coordinates of the targets in the fused target data are spatially matched with the three-dimensional region of interest to obtain matching results. Targets within the three-dimensional region of interest are filtered based on the matching results to obtain candidate target data. Invalid targets in the three-dimensional region of interest are removed based on the candidate target data, and the candidate target data is optimized to obtain potential collision target data. The relative distance and relative motion data between the vehicle and the potential target are calculated based on the potential collision target data to obtain motion parameter information. Spatial interference calculations are performed on the three-dimensional contour of the potential target through the three-dimensional region of interest to obtain collision point coordinates. The collision point velocity is calculated using the motion parameter information and the collision point coordinates. AI scene recognition is performed on the visual image data to obtain the traffic scene type; the initial collision time warning threshold and the initial collision time braking threshold are adjusted according to the traffic scene type to obtain the final collision time warning threshold and the final collision time braking threshold; the collision time is calculated using the collision point coordinates and collision point velocity; the collision time is analyzed using the final collision time warning threshold and the final collision time braking threshold; if the collision time does not exceed the final collision time braking threshold, the braking control signal is determined as a decision command and the decision command is executed; if the collision time does not exceed the final collision time warning threshold but exceeds the final collision time braking threshold, a warning is sent as a decision command and the decision command is executed.This invention solves the problem of inaccurate vehicle warnings and braking due to insufficient 3D perception of the inner wheel difference area during vehicle turning, enabling the detection of steering collisions and improving the accuracy of steering collision detection. Based on this invention, existing systems based on millimeter-wave radar or visual cameras have limited perception dimensions, lack 3D target information, struggle to obtain the true 3D contours and dimensions of obstacles, are prone to false alarms and missed alarms, and have fixed alarm thresholds, making them unsuitable for different traffic scenarios such as intersections and schools. These limitations lead to poor system protection and an inability to effectively avoid safety risks in the inner wheel difference area, resulting in low accuracy. Therefore, a steering collision detection method was designed, and its effectiveness was verified during steering collision detection. Finally, the accuracy of steering collision detection using this method was significantly improved.

[0025] In this embodiment, for ease of description, the steering collision detection device will be used as the execution subject in the following description.

[0026] Due to limitations in existing perception technologies and decision-making algorithms, the safety of large vehicle steering collision detection still needs improvement. Firstly, the perception dimension is limited, relying solely on millimeter-wave radar or visual cameras, lacking three-dimensional information and making it difficult to obtain the true outline and size of obstacles, easily leading to false alarms and missed alarms. Secondly, the risk area is static, using fixed monitoring areas, unable to dynamically match the true trajectory of the inner wheel difference with steering angle and vehicle speed, resulting in inaccurate detection. Thirdly, the decision threshold is fixed, using a uniform alarm threshold, unable to adapt to different scenarios such as intersections and schools, easily leading to false alarms or delayed responses. Fourthly, there is insufficient scenario adaptability; the model does not incorporate specialized learning of the inner wheel difference danger area, resulting in a significant decrease in protection effectiveness under complex road conditions, making it difficult to effectively protect the safety of vulnerable road users.

[0027] This application provides a solution that acquires 3D point cloud data and visual image data through a sensor module to achieve 3D environmental perception, making up for the lack of 2D perception. Then, it obtains accurate 3D target data through spatiotemporal fusion, generates a 3D area of ​​interest based on the driving state, matches the real inner wheel difference space, and obtains the accurate collision point and speed through spatial interference analysis. Based on this, it makes decisions on warning or braking, which solves the problem of inaccurate vehicle warning and braking caused by insufficient 3D perception of the inner wheel difference area when the vehicle is turning, and improves the accuracy of steering collision detection.

[0028] Based on this, embodiments of this application provide a steering collision detection method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the steering collision detection method of this application.

[0029] In this embodiment, the steering collision detection method is applied to a vehicle, and the steering collision detection method includes steps S01 to S04: Step S01: When the vehicle is detected to be turning, the vehicle driving status data, three-dimensional point cloud data and visual image data are acquired through the sensor module. Combination Figure 2 It is known that existing large vehicle turning safety protection systems based on millimeter-wave radar or visual cameras have limitations such as limited perception dimensions, static risk areas, and fixed alarm thresholds. They cannot accurately obtain three-dimensional information of obstacles, are difficult to adapt to real-time turning status and different traffic scenarios, and are prone to false alarms, missed alarms, and response delays. They cannot effectively avoid the risk of collision in the inner wheel difference area.

[0030] Therefore, this embodiment is designed for large vehicles (trucks, fire trucks, and dump trucks, etc.). To solve the above problems, when the vehicle is turning, the steering wheel angle signal is monitored for detection. When the steering signal is detected, the sensor module is immediately activated. The three-dimensional lidar performs real-time three-dimensional scanning of the vehicle's surrounding environment and analyzes the point cloud data. The fisheye camera acquires images of the environment on the turning side and performs preprocessing. The vehicle attitude sensor simultaneously acquires vehicle speed, steering wheel angle, and vehicle wheelbase. The three types of data are integrated simultaneously to form three-dimensional point cloud data, visual image data, and vehicle driving status data, respectively, providing basic data support for subsequent processing.

[0031] Step S02: The three-dimensional point cloud data and visual image data are spatiotemporally aligned and fused using a sensor fusion module to obtain fused target data. Subsequently, the 3D point cloud data and visual image data are time-stamped and synchronized to eliminate time discrepancies between the two types of data and ensure the synchronization of data acquisition. Then, based on the vehicle coordinate system, the two types of data are spatially transformed to unify the 3D point cloud data and visual image data under the same spatial reference system, eliminating spatial discrepancies. Based on the time synchronization results and spatial synchronization results, the feature information of the two types of data is extracted. Through feature fusion algorithms, complementary advantages are achieved, and fused target data with both 3D accuracy and image semantic information is integrated to improve the reliability and completeness of target detection.

[0032] Step S03: Generate a three-dimensional region of interest based on the vehicle driving status data, and filter the fused target data through the three-dimensional region of interest to obtain potential collision target data; Next, using the vehicle speed and steering wheel angle from the vehicle driving status data, the turning radius and inner wheel difference trajectory of the vehicle are calculated through the Ackerman steering dynamics model. Combined with the vehicle wheelbase, track width, and body height parameters, the two-dimensional inner wheel difference trajectory is extended to three-dimensional space to construct a three-dimensional region of interest that dynamically changes with the vehicle's steering state. Then, the three-dimensional coordinates of the targets in the fused target data are spatially matched with this three-dimensional region of interest to filter out candidate target data within the region, eliminate invalid targets, and optimize the data. Finally, accurate potential collision target data is obtained, reducing interference from irrelevant targets.

[0033] Step S04: Perform spatial interferometry analysis on the potential collision target data to obtain the collision point coordinates and collision point velocity, and generate and execute operation decisions based on the collision point coordinates and collision point velocity.

[0034] Finally, the potential collision target data is preprocessed to calculate the relative distance and relative motion data between the vehicle and the potential target, obtain motion parameter information, analyze the spatial interference relationship between the three-dimensional contour of the potential target and the three-dimensional area of ​​interest, accurately calculate the collision point coordinates, and combine the motion parameter information and the collision point coordinates to further calculate the collision point velocity. AI scene recognition is performed through visual image data to determine the traffic scene type and dynamically adjust the collision time warning and braking thresholds. The collision time is calculated and compared with the dynamic threshold. Based on the comparison result, a graded warning or active braking command is generated and sent to the vehicle actuator to complete the execution.

[0035] The basic architecture of this embodiment is as follows: Figure 3 As shown, the overall system architecture is centered on the domain controller, integrating multi-sensor fusion algorithms, Ackerman dynamics models, and real-time decision-making logic. The sensor module consists of fisheye cameras, 3D LiDAR, and vehicle attitude sensors, which respectively collect visual image data, 3D point cloud data, and vehicle driving status data, and transmit them to the domain controller.

[0036] The domain controller first completes the spatiotemporal alignment and fusion of two types of perception data through the sensor fusion module, then generates a three-dimensional region of interest based on the driving status data and filters the target, then performs spatial interference analysis to obtain collision-related parameters, and finally generates operation decisions and sends them to the brake execution controller, the vehicle instrument panel and the vehicle central control screen and other actuators to realize the closed loop of the entire process of steering collision detection and protection.

[0037] Specifically, the aforementioned sensor module includes a 3D LiDAR, a fisheye camera, and a vehicle attitude sensor. Therefore, the steps for acquiring vehicle driving status data, 3D point cloud data, and visual image data through the sensor module include: Step S011: The three-dimensional LiDAR is used to perform three-dimensional scanning and point cloud analysis of the vehicle's surrounding environment to obtain three-dimensional point cloud data; Step S012: The fisheye camera is used to acquire and preprocess images of the vehicle's steering side environment to obtain visual image data; Step S013: The vehicle speed, steering wheel angle and wheelbase are collected by the vehicle attitude sensor, and the vehicle speed, steering wheel angle and wheelbase are integrated to obtain vehicle driving status data.

[0038] The 3D LiDAR in this embodiment features high point cloud density and fast scanning capabilities. It connects to the domain controller via an in-vehicle Ethernet network, with a point cloud transmission rate of no less than 10Hz. During operation, it performs real-time 3D scanning of the vehicle's surroundings, especially the steering side area, capturing the 3D contour information of various targets in the environment. The raw point cloud data obtained from the scan is then processed through decoding, noise reduction, and filtering to remove outliers and abnormal data, extracting effective point cloud information, and ultimately forming high-precision 3D point cloud data, providing support for target 3D contour reconstruction and spatial positioning.

[0039] The fisheye camera is specifically designed for target recognition and classification in turning-side scenes. It features starlight-level night vision and can operate stably under various lighting conditions. It connects directly to the domain controller via an AHD interface with a transmission latency of no more than 150ms and an AI detection frame rate of no less than 25FPS. After acquiring the original color image of the turning-side environment, it performs preprocessing operations such as image format conversion, color restoration, distortion correction, noise reduction, and enhancement to eliminate image distortion and noise interference, improve image clarity, and provide high-quality image input for subsequent AI target detection and scene recognition.

[0040] Vehicle attitude sensors collect core parameters such as vehicle speed, steering wheel angle, and wheelbase in real time during vehicle operation. After collection, various parameters are verified and standardized to remove abnormal values. Through data integration algorithms, the three types of parameters are correlated and fused to form complete and accurate vehicle driving status data, providing accurate vehicle physical parameter support for Ackerman steering dynamics model calculation and three-dimensional region of interest generation.

[0041] More specifically, step S02 above, which involves spatiotemporally aligning and fusing the 3D point cloud data and visual image data using a sensor fusion module to obtain the fused target data, includes: Step S021: Perform timestamp synchronization processing on the three-dimensional point cloud data and visual image data to obtain time synchronization results; Step S022: Perform spatial coordinate transformation on the three-dimensional point cloud data and visual image data according to the coordinate system of the vehicle to obtain spatial synchronization results; Step S023: Based on the spatial synchronization result and the temporal synchronization result, feature fusion is performed on the three-dimensional point cloud data and visual image data to obtain fused target data.

[0042] During timestamp synchronization, the acquisition timestamps of the 3D point cloud data and visual image data are extracted. Based on the unified clock reference of the domain controller, a time interpolation algorithm is used to calibrate the two types of data to eliminate time deviations caused by different sensor acquisition frequencies. This ensures that the point cloud data and image data acquired at the same time correspond accurately, resulting in time synchronization.

[0043] During the spatial coordinate transformation stage, the intrinsic and extrinsic calibration parameters of the 3D LiDAR and fisheye camera are first obtained through the LiDAR calibration module and the image calibration module, respectively. Based on the vehicle coordinate system (X-axis forward, Y-axis left, origin at the rear axle center), the 3D point cloud data is transformed from the LiDAR coordinate system and the visual image data is transformed from the camera coordinate system to the vehicle coordinate system, respectively, to achieve spatial unification of the two types of data and obtain spatial synchronization results.

[0044] In the feature fusion process, spatial information such as the target's 3D coordinates and contour features are extracted from the 3D point cloud data, and semantic information such as the target's color, texture, and category are extracted from the visual image data. A feature-level fusion algorithm is used to associate and integrate the two types of feature information, making up for the shortcomings of a single sensor, filtering redundant information, and strengthening effective features. Finally, fused target data with both spatial accuracy and semantic information is generated, improving the accuracy and completeness of target detection.

[0045] Further, step S03 above, the step of generating a three-dimensional region of interest based on the vehicle driving state data, includes: Step S031: Based on the vehicle speed and steering wheel angle in the vehicle driving status data, calculate the vehicle turning radius and inner wheel difference trajectory using the Ackermann steering dynamics model to obtain the inner wheel difference trajectory information. Step S032: Based on the wheelbase, track width, and vehicle height parameters of the vehicle, the inner wheel difference trajectory is extended from a two-dimensional plane to a three-dimensional space to obtain spatial trajectory information; Step S033: Construct a three-dimensional region of interest based on the spatial trajectory information.

[0046] like Figure 4 As shown, firstly, based on the steering wheel angle and steering ratio in the vehicle driving status data, the formula is used... Calculate the actual wheel rotation angle, where, The steering wheel angle (degrees). The steering gear ratio is dimensionless. The actual wheel turning angle (degrees) is then converted to radians. Combined with vehicle speed and wheelbase, the front wheel turning radius is calculated using the Ackermann steering dynamics model. and rear wheel turning radius Then, the two-dimensional trajectory of the inner wheel difference during vehicle turning is fitted to obtain the inner wheel difference trajectory information, where, .

[0047] Subsequently, the vehicle body parameters in the vehicle calibration module are called to obtain data such as vehicle wheelbase, track width, and vehicle height. Based on the two-dimensional inner wheel difference trajectory, the trajectory is expanded in the vertical direction by combining vehicle height and wheel radius, thus elevating the two-dimensional trajectory to a three-dimensional spatial trajectory. The three-dimensional trajectories of the upper right corner of the front of the vehicle and the center point of the right rear wheel are given special consideration. The formula for calculating the upper right corner of the vehicle's front is: Formula ① , formula② And formula ③ Calculate the top corner point of the front of the vehicle, where, Angular velocity is the parameter that reflects the dynamic characteristics of the motion, and t is the prediction time variable, corresponding to the TTC (Time to Collision) time series. The vehicle's height is calibrated to ensure the accuracy of the trajectory's height dimension.

[0048] The formula for calculating the center point of the right rear wheel is: Formula ① , formula② , formula③ .

[0049] The spatial trajectory information is composed of the three-dimensional trajectories of the upper right corner of the vehicle's front and the center point of the right rear wheel. Finally, based on this spatial trajectory information, a closed three-dimensional geometric modeling method is used to construct a closed, dynamically changing three-dimensional area of ​​interest that changes in real time with the vehicle's steering state (vehicle speed, steering angle). This area not only covers the inner wheel difference area on the ground but also includes the air area, enabling comprehensive monitoring of both ground and air obstacles. Figure 2 The red area shown is the trajectory predicted based on the input vehicle parameters. This trajectory has an altitude, which is adjusted according to the vehicle's height. Through this 3D stereoscopic focus area, it is possible to provide early warning and braking not only for ground targets but also for aerial targets (such as tree branches and road signs). It can also effectively filter targets taller than the vehicle, and will not cause false alarms when passing through height restriction poles or tunnels.

[0050] Furthermore, step S03 above, which filters the fused target data through the three-dimensional region of interest to obtain potential collision target data, includes: Step S034: Spatial matching is performed between the three-dimensional coordinates of the target in the fused target data and the three-dimensional region of interest to obtain the matching result; Step S035: Based on the matching results, the targets within the three-dimensional area of ​​interest are filtered to obtain candidate target data; Step S036: Eliminate invalid targets in the three-dimensional region of interest based on the candidate target data, and optimize the candidate target data to obtain potential collision target data.

[0051] In combination with the above Figure 4 After obtaining the three-dimensional region of interest, the spatial boundary parameters of the constructed three-dimensional region of interest are retrieved. These parameters include core information such as the three-dimensional coordinate range of the region, the boundary contour equation, and the dynamic update frequency, ensuring that they are completely matched with the inner wheel difference trajectory under the current vehicle steering state.

[0052] Subsequently, the complete three-dimensional coordinate information (X, Y, and Z axis coordinates, with precision retained to the millimeter level) of each target in the fused target data is extracted and precisely geometrically compared with the spatial range of the three-dimensional region of interest. An inclusion judgment algorithm for spatial points and closed regions is adopted (specifically, the ray method, which involves emitting rays in any direction and counting the number of intersections between the rays and the region boundary; if the number of intersections is odd, the target coordinates are determined to be within the region, and if it is even, it is determined to be outside the region). The three-dimensional coordinates of each target are verified one by one to see if they fall within the spatial boundary of the three-dimensional region of interest. The coordinate information, matching status (inclusion or non-inclusion), and matching confidence of each target are recorded in detail to form a complete spatial matching result. Targets with a matching confidence of less than 95% are temporarily listed as suspicious targets and will be verified again later.

[0053] Based on the matching results, all targets whose three-dimensional coordinates are within the three-dimensional area of ​​interest and whose matching confidence is ≥95% are selected and integrated to form candidate target data. Irrelevant targets outside the area and suspicious targets with insufficient matching confidence are simultaneously removed, which effectively reduces the amount of data for subsequent spatial interferometry analysis, reduces the probability of false alarms, and improves data processing efficiency.

[0054] The candidate target data undergoes deep optimization processing. First, an outlier detection algorithm (using the 3σ principle) is used to remove invalid targets whose coordinate offset exceeds a preset threshold or whose contour incompleteness is ≥30%. Then, a Kalman filter algorithm (with the state equation as follows) is applied. The observation equation is ,in The target state vector contains three-dimensional coordinates and velocity. and The remaining candidate targets' 3D coordinates and motion states are smoothed by combining process noise and observation noise (respectively). This compensates for deviations caused by sensor accuracy and environmental interference during data acquisition, supplements missing target feature information (such as target motion acceleration and contour details), and associates target category information from visual image data. It also performs secondary verification of candidate targets by combining the target's shape and texture features, distinguishing between different types of targets such as pedestrians, non-motorized vehicles, and small vehicles, ensuring the accuracy of target recognition. Ultimately, accurate and reliable potential collision target data is obtained, providing high-quality target data support for subsequent spatial interferometry analysis and collision point calculation, and ensuring the accuracy and response speed of subsequent collision detection.

[0055] This embodiment, through the above-described scheme, specifically acquires vehicle driving state data, 3D point cloud data, and visual image data via a sensor module when the vehicle is turning. A sensor fusion module performs spatiotemporal alignment and fusion of the 3D point cloud data and visual image data to obtain fused target data. A 3D region of interest is generated based on the vehicle driving state data, and the fused target data is filtered through this region of interest to obtain potential collision target data. Spatial interference analysis is performed on the potential collision target data to obtain the collision point coordinates and velocity. An operational decision is then generated and executed based on these collision point coordinates and velocity. Thus, by acquiring 3D point cloud data and visual image data through the sensor module, 3D environmental perception is achieved, compensating for the lack of 2D perception. Subsequently, accurate 3D target data is obtained through spatiotemporal fusion. A 3D region of interest is generated based on the driving state, matching the actual inner wheel difference space. Spatial interference analysis then yields the precise collision point and velocity, enabling decision-making for warning or braking. This solves the problem of inaccurate vehicle warnings and braking due to insufficient 3D perception of the inner wheel difference area during vehicle turning, improving the accuracy of steering collision detection.

[0056] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 5 In step S04, where spatial interferometry analysis is performed on the potential collision target data to obtain the collision point coordinates and collision point velocity, the steering collision detection method further includes steps S041-S043: Step S041: Calculate the relative distance and relative motion data between the vehicle and the potential target based on the potential collision target data to obtain motion parameter information; Step S042: Perform spatial interference calculation on the three-dimensional contour of the potential target through the three-dimensional region of interest to obtain the coordinates of the collision point; Step S043: The velocity at the collision point is calculated using the motion parameter information and the coordinates of the collision point.

[0057] First, the potential collision target data is systematically standardized and preprocessed to ensure the uniformity, validity, and accuracy of the data, providing a reliable data foundation for subsequent calculations.

[0058] Secondly, the three-dimensional coordinates of all potential collision targets are uniformly transformed to the vehicle coordinate system (X-axis forward, Y-axis left, origin at the rear axle center). Through a preset coordinate transformation matrix, coordinate deviations when data is collected from different sensors are eliminated, ensuring that all target coordinates are in the same spatial reference system, which facilitates the subsequent calculation of relative distance and relative motion parameters. At the same time, the timestamps of all potential collision target data and vehicle driving status data are synchronized. Based on the unified clock reference of the domain controller, a linear interpolation algorithm is used to calibrate data with timestamp deviations, ensuring that the real-time vehicle status and target data are completely synchronized, avoiding calculation errors caused by time misalignment.

[0059] Subsequently, a 3σ outlier detection algorithm was used to filter the preprocessed potential collision target data. A data deviation threshold was set, and outliers and invalid data with coordinate offsets exceeding the threshold, abnormal motion state parameters, etc., were removed. Simultaneously, missing target feature data was interpolated to further ensure data integrity and validity. After preprocessing, the motion parameter information calculation stage began. First, the three-dimensional coordinates (X, Y, Z) of each potential collision target and the real-time position coordinates of the vehicle itself were extracted. Then, the spatial Euclidean distance formula (…) was used… It accurately calculates the real-time relative distance between the vehicle and each potential collision target, retaining three decimal places of precision to ensure the accuracy of distance calculation.

[0060] Next, combining multiple frames of potential collision target data, the motion trajectory of each target is tracked. A combination of frame differencing and Kalman filtering is used to estimate the relative motion parameters between the target and the vehicle. Specifically, frame differencing calculates the change in target coordinates between adjacent frames to obtain a rough estimate of the relative velocity. Then, the Kalman filtering algorithm (with the state equation as follows) is applied... The observation equation is ,in The target state vector contains three-dimensional coordinates and velocity. and The rough velocity value is smoothed by separating process noise and observation noise (with the noise variance preset to a fixed value), eliminating fluctuations caused by environmental interference and sensor errors. At the same time, key parameters such as relative motion acceleration and motion direction angle are estimated. The parameters such as relative distance, relative motion velocity, relative motion acceleration, and motion direction angle are integrated and summarized to form complete motion parameter information, providing comprehensive parameter support for the subsequent calculation of collision point coordinates and collision point velocity.

[0061] Next, the collision point coordinate calculation stage begins. The spatial boundary parameters of the previously constructed 3D region of interest (including the region's 3D coordinate range, boundary contour equation, dynamic update frequency, etc.) and the 3D contour data of each potential collision target (composed of target contour point clouds extracted from the fused target data) are retrieved. A spatial interferometry detection algorithm (specifically, the separating axis theorem) is used to comprehensively analyze the spatial relationship between the target's 3D contour and the 3D region of interest, determining whether there is a spatial intersection. If an intersection exists, the coordinates of the center point of the intersection region are further calculated. This center point is the collision point where the vehicle may collide with the target during steering. Through precise spatial coordinate calculation, the 3D coordinates of the collision point (X-collision, Y-collision, Z-collision) are obtained with millimeter-level accuracy, ensuring the accuracy of collision point positioning.

[0062] Finally, the velocity at the collision point is calculated. Combining the motion parameters and collision point coordinates obtained above, the kinematic vector decomposition formula is used to decompose the relative velocity into velocity components along the normal and tangential directions of the collision point. The velocity component along the normal direction of the collision point is the core component of the collision point velocity. The velocity components are then dynamically corrected by the relative acceleration to eliminate the influence of velocity changes caused by acceleration. The accurate velocity value of the collision point is obtained by integrating the calculations. At the same time, the direction parameters of the collision point velocity are recorded to form complete collision point velocity information. This provides accurate quantitative basis for subsequent collision time (TTC) calculation, risk level determination, and decision analysis, ensuring the scientific nature and reliability of subsequent decision instructions.

[0063] Specifically, step S04 above, which involves generating and executing an operation decision based on the collision point coordinates and collision point velocity, includes: Step S044: Perform AI scene recognition using the visual image data to obtain the traffic scene type; Step S045: Adjust the initial collision time warning threshold and the initial collision time braking threshold according to the traffic scenario type to obtain the final collision time warning threshold and the final collision time braking threshold. Step S046: The collision time is calculated using the collision point coordinates and the collision point velocity; Step S047: Perform decision analysis on the collision time using the final collision time warning threshold and the final collision time braking threshold; Step S048: If the collision time does not exceed the final collision time braking threshold, then the braking control signal is determined as a decision command, and the decision command is executed. Step S049: If the collision time does not exceed the final collision time warning threshold and exceeds the final collision time braking threshold, then a warning will be sent as a decision instruction and the decision instruction will be executed.

[0064] It should be clear that the AI ​​scene recognition in this embodiment uses an AI model, and the training of the AI ​​model is as follows: Figure 6 As shown, a panoramic camera system is used to systematically collect a large number of images and video sequences covering different time periods (day / night), different weather conditions (sunny / rainy / foggy), different lighting conditions (front lighting / backlighting / shade), and various typical urban scenes, which are used as the raw training data.

[0065] Subsequently, a refined annotation system was constructed to perform 2D bounding boxes and category annotations on dynamic and static obstacles in the images. Categories included pedestrians, bicycles, electric bicycles, motorcycles, small passenger cars, and large vehicles. Simultaneously, semantic segmentation or keypoint annotation was performed on fixed traffic elements in the image background, including school zone signs, yield signs, pedestrian crossings, stop lines, traffic lights (and their status), and road turning arrows. The collected data underwent standardized preprocessing, including resolution unification, format conversion, and outlier frame removal, and was divided into training, validation, and test sets. Training was performed using a neural network with two parallel task heads. The object detection head outputs the bounding boxes, category confidence scores, and category labels for all objects in the image, while the scene recognition head outputs scene classification labels or pixel-level positions of specific traffic elements.

[0066] The trained model is transformed and optimized by means of model pruning, quantization, knowledge distillation, etc., to reduce model complexity and computational load while maintaining accuracy as much as possible. Finally, it is converted into a format supported by the embedded hardware platform inference engine to complete the training of the AI ​​detection module.

[0067] After training the AI ​​detection module, AI scene recognition can be performed, such as... Figure 7As shown, the visual image data acquired and preprocessed by the fisheye camera is input into the trained AI detection module in real time for forward inference. Each frame of the image outputs structured information, including a list of targets. Each target contains its bounding box in 2D pixel coordinates, target type, and presence confidence. Through pre-completed camera intrinsic and extrinsic parameter calibration, combined with the target's position in the image and ground assumptions or stereo vision principles, the distance of the target relative to the vehicle is calculated. At the same time, scene semantic information is output, including the scene category of the current frame (such as school road, traffic light intersection, ordinary road, etc.) and its confidence, or the recognition results of key traffic signs / markings are directly output.

[0068] The same target in consecutive frames is tracked in association, and its relative speed is estimated using algorithms such as frame difference method or Kalman filter. Finally, the current traffic scene type is determined, which provides a basis for subsequent threshold adjustment.

[0069] After obtaining the recognition results, corresponding threshold calculations and decision analyses can be performed, such as... Figure 8 As shown, the system first receives the scene classification results output by the AI ​​detection module, verifies the classification confidence (threshold ≥ 0.8), and simultaneously statistically analyzes the target type distribution, including scene features such as the number of pedestrians, the number of vehicles, and the proportion of vulnerable road users. Based on the scene features, a dynamic mapping of three levels of TTC thresholds is performed to establish the correspondence between scene features and TTC thresholds. The first-level TTC threshold (lenient) is used for low-risk scenes, the second-level TTC threshold (medium) is used for medium-risk scenes, and the third-level TTC threshold (strict) is used for high-risk scenes.

[0070] Threshold fine-tuning is performed based on multiple factors. Regarding the pedestrian density weighting coefficient, when the number of pedestrians in the detection area is greater than 10, the threshold at each level is reduced by 0.2 seconds, and for every additional 5 pedestrians, it is reduced by an additional 0.1 seconds (maximum adjustment of 0.5 seconds). Regarding the traffic sign influence coefficient, the threshold at each level is reduced by 0.3 seconds when a yield sign is identified, 0.2 seconds when a turning warning sign is identified, and 0.3 seconds when a school zone sign is identified. Regarding the intersection type influence coefficient, the threshold at each level is reduced by 0.3 seconds when a traffic light-controlled intersection is identified, and 0.5 seconds when a pedestrian crossing is identified. When no marked scene is identified, the default inner wheel difference alarm TTC and braking TTC are used to obtain the final collision time warning threshold and the final collision time braking threshold.

[0071] Simultaneously, using the collision point coordinates and collision point velocity, the formula is... Calculate the collision time, where, , , ,and Calculate the minimum TTC for each objective, identify the most pressing objective, compare the minimum TTC with the final threshold, and complete the decision analysis.

[0072] Finally, the decision is implemented. Figure 9 As shown, the decision analysis results are sent to the state machine for debouncing processing. Multi-frame judgment processing is used to determine whether to switch to the alarm state, effectively filtering out instantaneous false alarms.

[0073] If the collision time does not exceed the final collision time braking threshold, it indicates an extremely high collision risk. The braking control signal is then identified as a decision command and sent to the vehicle's brake execution controller to initiate active braking, reduce vehicle speed, or stop the vehicle to avoid a collision.

[0074] If the collision time does not exceed the final collision time warning threshold but exceeds the final collision time braking threshold, it indicates that there is a collision risk but the emergency braking conditions have not been met. A warning will be sent as a decision command, and the vehicle's instrument panel and central control screen will output audible and visual warning signals to remind the driver to adjust the steering angle or vehicle speed in time to avoid the collision risk.

[0075] If the collision time exceeds the final collision time warning threshold, it indicates that there is no collision risk, no instructions are output, and the vehicle continues to drive normally.

[0076] This embodiment, through the above-described scheme, specifically calculates the relative distance and relative motion data between the vehicle and the potential target based on the potential collision target data to obtain motion parameter information; performs spatial interference calculations on the three-dimensional contour of the potential target using the three-dimensional region of interest to obtain the collision point coordinates; and calculates the collision point velocity using the motion parameter information and the collision point coordinates. Thus, by acquiring three-dimensional point cloud data and visual image data through the sensor module, three-dimensional environmental perception is achieved, compensating for the lack of two-dimensional perception. Subsequently, accurate three-dimensional target data is obtained through spatiotemporal fusion, and a three-dimensional region of interest is generated based on the driving state, matching the actual inner wheel difference space. Then, through spatial interference analysis, the precise collision point and velocity are obtained, and a decision to warn or brake is made accordingly. This solves the problem of inaccurate vehicle warnings and braking due to insufficient three-dimensional perception of the inner wheel difference area during vehicle turning, improving the accuracy of steering collision detection.

[0077] For example, to help understand the implementation flow of the steering collision detection method obtained by combining this embodiment with the above embodiment one, please refer to... Figure 10 , Figure 10 A simplified flowchart of a steering collision detection method is provided, specifically: First, after the vehicle starts, the sensor module (3D LiDAR, fisheye camera, vehicle attitude sensor) remains on standby. When the vehicle starts to turn, data acquisition is immediately initiated to acquire 3D point cloud data, visual image data, and vehicle driving status data.

[0078] Subsequently, the sensor fusion module performs time stamp synchronization and spatial coordinate transformation on the 3D point cloud data and visual image data to achieve spatiotemporal alignment, and then obtains the fused target data through feature fusion.

[0079] Next, based on the vehicle driving status data, the inner wheel difference trajectory is calculated using the Ackermann steering dynamics model, extended to three-dimensional space, and a three-dimensional region of interest is constructed. The target data is then filtered and fused using this region to obtain potential collision target data.

[0080] Then, spatial interferometry analysis is performed on the potential collision target data to calculate the relative motion parameters between the vehicle and the target, the coordinates of the collision point, and the velocity of the collision point. At the same time, AI scene recognition is performed through visual image data to determine the traffic scene type and dynamically adjust the TTC threshold. The collision time is calculated and compared with the dynamic threshold to generate hierarchical decision instructions.

[0081] Finally, the decision command is sent to the vehicle actuator to perform a warning or active braking operation, completing the entire steering collision detection and protection process. The entire process achieves dynamic perception, accurate decision-making, and rapid response, effectively avoiding the collision risk in the inner wheel difference area.

[0082] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the steering collision detection method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0083] This application also provides a steering collision detection device, please refer to... Figure 11 The steering collision detection device is applied to a vehicle, and the steering collision detection device includes: The acquisition module 10 is used to acquire vehicle driving status data, three-dimensional point cloud data and visual image data through the sensor module when the vehicle is detected to be turning. The fusion module 20 is used to perform spatiotemporal alignment and fusion of the three-dimensional point cloud data and visual image data through the sensor fusion module to obtain fused target data; The filtering module 30 is used to generate a three-dimensional region of interest based on the vehicle driving status data, and to filter the fused target data through the three-dimensional region of interest to obtain potential collision target data. The execution module 40 is used to perform spatial interference analysis on the potential collision target data to obtain the collision point coordinates and collision point velocity, and to generate and execute operation decisions based on the collision point coordinates and collision point velocity.

[0084] The steering collision detection device provided in this application, employing the steering collision detection method described in the above embodiments, can solve the technical problem of inaccurate vehicle warning and braking due to insufficient three-dimensional perception of the inner wheel difference area during vehicle turning. Compared with the prior art, the beneficial effects of the steering collision detection device provided in this application are the same as those of the steering collision detection method provided in the above embodiments, and other technical features in the steering collision detection device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0085] This application provides a steering collision detection device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steering collision detection method in the first embodiment described above.

[0086] The following is for reference. Figure 12 The diagram illustrates a structural schematic suitable for implementing the steering collision detection device in the embodiments of this application. The steering collision detection device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 12 The steering collision detection device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of this application.

[0087] like Figure 12As shown, the steering collision detection device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the steering collision detection device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. The communication device 1009 allows the steering collision detection device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows a steering collision detection device with various systems, it should be understood that it is not required to implement or have all of the systems shown. More or fewer systems may be implemented alternatively.

[0088] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0089] The steering collision detection device provided in this application, employing the steering collision detection method described in the above embodiments, can solve the technical problem of inaccurate vehicle warning and braking due to insufficient three-dimensional perception of the inner wheel difference area during vehicle turning. Compared with the prior art, the beneficial effects of the steering collision detection device provided in this application are the same as those of the steering collision detection method provided in the above embodiments, and other technical features of this steering collision detection device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0090] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0091] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0092] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the steering collision detection method in the above embodiments.

[0093] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0094] The aforementioned computer-readable storage medium may be included in the steering collision detection device; or it may exist independently and not be assembled into the steering collision detection device.

[0095] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by the steering collision detection device, the steering collision detection device: when detecting that the vehicle is turning, acquires vehicle driving state data, 3D point cloud data, and visual image data through a sensor module; performs spatiotemporal alignment and fusion of the 3D point cloud data and visual image data through a sensor fusion module to obtain fused target data; generates a 3D region of interest based on the vehicle driving state data, and filters the fused target data through the 3D region of interest to obtain potential collision target data; performs spatial interference analysis on the potential collision target data to obtain collision point coordinates and collision point velocity, and generates and executes an operation decision based on the collision point coordinates and collision point velocity.

[0096] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0097] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0098] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0099] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described steering collision detection method. This solves the technical problem of inaccurate vehicle warning and braking due to insufficient three-dimensional perception of the inner wheel difference area during vehicle turning. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the steering collision detection method provided in the above embodiments, and will not be elaborated upon here.

[0100] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the steering collision detection method described above.

[0101] The computer program product provided in this application can solve the technical problem of inaccurate vehicle warning and braking due to insufficient three-dimensional perception of the inner wheel difference area during vehicle turning. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the steering collision detection method provided in the above embodiments, and will not be repeated here.

[0102] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A steering collision detection method, characterized in that, The steering collision detection method is applied to a vehicle, and the steering collision detection method includes: When the vehicle is detected to be turning, the sensor module acquires vehicle driving status data, 3D point cloud data, and visual image data. The sensor fusion module performs spatiotemporal alignment and fusion of the 3D point cloud data and visual image data to obtain fused target data. A three-dimensional region of interest is generated based on the vehicle driving status data, and the fused target data is filtered through the three-dimensional region of interest to obtain potential collision target data. Spatial interferometry analysis is performed on the potential collision target data to obtain the collision point coordinates and collision point velocity. Operation decisions are generated and executed based on the collision point coordinates and collision point velocity.

2. The steering collision detection method as described in claim 1, characterized in that, The sensor module includes a 3D LiDAR, a fisheye camera, and a vehicle attitude sensor. The steps of acquiring vehicle driving status data, 3D point cloud data, and visual image data through the sensor module include: The three-dimensional lidar is used to perform three-dimensional scanning and point cloud analysis of the vehicle's surrounding environment to obtain three-dimensional point cloud data. The fisheye camera is used to acquire and preprocess images of the vehicle's steering side environment to obtain visual image data. The vehicle speed, steering wheel angle, and wheelbase are collected by the vehicle attitude sensor, and the vehicle driving status data are obtained by integrating the vehicle speed, steering wheel angle, and wheelbase.

3. The steering collision detection method as described in claim 1, characterized in that, The step of performing spatiotemporal alignment and fusion of the 3D point cloud data and visual image data using a sensor fusion module to obtain the fused target data includes: The three-dimensional point cloud data and visual image data are time-stamped and synchronized to obtain a time synchronization result. Based on the vehicle's coordinate system, the three-dimensional point cloud data and visual image data are transformed into spatial coordinates to obtain spatial synchronization results; Based on the spatial synchronization results and temporal synchronization results, feature fusion is performed on the 3D point cloud data and visual image data to obtain fused target data.

4. The steering collision detection method as described in claim 2, characterized in that, The step of generating a three-dimensional region of interest based on the vehicle driving status data includes: Based on the vehicle speed and steering wheel angle in the vehicle driving status data, the turning radius and inner wheel difference trajectory of the vehicle are calculated using the Ackermann steering dynamics model to obtain the inner wheel difference trajectory information; Based on the vehicle's wheelbase, track width, and body height parameters, the inner wheel difference trajectory is extended from a two-dimensional plane to a three-dimensional space to obtain spatial trajectory information; A three-dimensional region of interest is constructed based on the spatial trajectory information.

5. The steering collision detection method as described in claim 4, characterized in that, The step of filtering the fused target data through the three-dimensional region of interest to obtain potential collision target data includes: The target's three-dimensional coordinates in the fused target data are spatially matched with the three-dimensional region of interest to obtain the matching result; Based on the matching results, targets within the three-dimensional area of ​​interest are filtered to obtain candidate target data; Invalid targets in the three-dimensional region of interest are removed based on the candidate target data, and the candidate target data is optimized to obtain potential collision target data.

6. The steering collision detection method as described in claim 1, characterized in that, The step of performing spatial interferometry analysis on the potential collision target data to obtain the collision point coordinates and collision point velocity includes: The relative distance and relative motion data between the vehicle and the potential target are calculated based on the potential collision target data to obtain motion parameter information; Spatial interference calculations are performed on the three-dimensional contour of the potential target through the three-dimensional region of interest to obtain the collision point coordinates; The velocity at the collision point is calculated using the motion parameters and the coordinates of the collision point.

7. The steering collision detection method as described in claim 1, characterized in that, The step of generating and executing an operation decision based on the collision point coordinates and collision point velocity includes: The traffic scene type is obtained by performing AI scene recognition on the visual image data. The initial collision time warning threshold and the initial collision time braking threshold are adjusted according to the traffic scenario type to obtain the final collision time warning threshold and the final collision time braking threshold. The collision time is calculated using the collision point coordinates and the collision point velocity. The collision time is analyzed for decision-making based on the final collision time warning threshold and the final collision time braking threshold. If the collision time does not exceed the final collision time braking threshold, the braking control signal is determined as a decision command, and the decision command is executed. If the collision time does not exceed the final collision time warning threshold but exceeds the final collision time braking threshold, a warning will be sent as a decision instruction and the decision instruction will be executed.

8. A steering collision detection device, characterized in that, The steering collision detection device is applied to a vehicle, and the steering collision detection device includes: The acquisition module is used to acquire vehicle driving status data, 3D point cloud data and visual image data through the sensor module when the vehicle is detected to be turning. The fusion module is used to perform spatiotemporal alignment and fusion of the 3D point cloud data and visual image data through the sensor fusion module to obtain fused target data. The filtering module is used to generate a three-dimensional region of interest based on the vehicle driving status data, and to filter the fused target data through the three-dimensional region of interest to obtain potential collision target data. The execution module is used to perform spatial interferometry analysis on the potential collision target data to obtain the collision point coordinates and collision point velocity, and to generate and execute operation decisions based on the collision point coordinates and collision point velocity.

9. A steering collision detection device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the steering collision detection method as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the steering collision detection method as described in any one of claims 1 to 7.

Images