A collision type identification method, system and device based on collision point prediction

Abstract

The application provides a collision type identification method, system and device based on collision point prediction, which synchronizes, compensates and weights the multi-source perception data of a target vehicle in time and space to obtain a target feature vector; based on the target feature vector and state data, collision point information is predicted to determine the collision type and the corresponding confidence; based on the millimeter wave radar and the visual sensor, a normalized consistency score is calculated, and when the normalized consistency score is lower than a preset threshold, it is degraded to a basic safety mode; based on the collision type and the corresponding confidence, a safety control strategy is generated; the target feature vector is constructed by fusing multi-source perception data to achieve high-precision perception, and by introducing a random forest classifier and a weighted confidence mechanism of collision point uncertainty measurement, the system can realize hierarchical response when part of the sensor information is missing, thereby improving the robustness of decision-making and the adaptability to different collision scenarios under complex working conditions.

Description

Technical Field

[0001] This application relates to the field of vehicle active safety technology, specifically to a collision type identification method, system, and device based on collision point prediction. Background Technology

[0002] With the continuous improvement of automotive intelligence, pre-collision safety systems have become one of the core safety functions standard in modern vehicles. These systems typically rely on onboard environmental perception sensors such as millimeter-wave radar, vision cameras, and lidar to detect obstacles in front of the vehicle in real time and calculate the time to collision (TTC) based on motion parameters such as relative distance and relative speed. When the system determines that a collision is unavoidable, it triggers a series of passive safety measures, including automatic emergency braking (AEB), seatbelt pretensioning, and airbag deployment, to reduce the risk of occupant injury during the collision. The timing of the pre-collision system's activation and the type of collision have a significant impact on the occupant's posture, position before the collision, and the final risk of injury. Evaluating pre-collision safety strategies from the perspective of whole-vehicle system integration has become an important direction for improving vehicle safety performance.

[0003] However, existing pre-collision safety systems still have the following technical limitations in practical applications: First, the system lacks sufficient ability to predict collision types, resulting in a lack of differentiated protection strategies. Existing systems rely solely on scalar indicators such as TTC (Total Collision Trace) for a binary judgment of "whether a collision has occurred," failing to effectively predict "how a collision will occur." Different collision types (such as 100% frontal overlap, 40% frontal offset, 25% small overlap, and oblique collisions) cause fundamentally different injury mechanisms to occupants. For example, in a 100% frontal overlap collision, the impact force is uniformly transmitted to the energy-absorbing structure at the front of the vehicle, with occupants primarily experiencing longitudinal deceleration; however, in a small overlap collision, the impact force bypasses the longitudinal beams and acts directly on the passenger compartment, potentially leading to structural intrusion into the passenger compartment and significant lateral displacement of the occupants. Existing systems employ a general triggering method, making it difficult to implement differentiated protection strategies based on specific collision types.

[0004] Second, the fusion of multi-sensor information is insufficient, resulting in low accuracy in predicting collision points and angles. Millimeter-wave radar can provide high-precision distance and velocity information, but its angular resolution is limited; visual sensors can identify semantic information such as target category, orientation, and lane lines, but are easily affected by environmental factors such as lighting and weather; while lidar can provide dense 3D point cloud data, its computational complexity is high and its cost is high. Existing systems lack an effective multi-sensor fusion mechanism, making it difficult to accurately acquire key geometric information such as the relative attitude of the target vehicle and the vehicle itself, the expected contact point, and the collision angle, thus limiting the ability to make refined judgments about collision scenarios.

[0005] Third, rigid safety response strategies are difficult to adapt to complex collision scenarios. Current pre-collision systems rely heavily on pre-calibrated rules for their safety response logic, lacking adaptability to different collision scenarios. Typical problems include: in 25% small overlap collisions, the front impact sensor (FIS) may fail to detect sufficient impact signals because the collision force does not travel through the traditional force transmission path, leading to delayed airbag deployment; in oblique collisions, the ideal protection scheme should activate both front and side airbags simultaneously, but existing systems often only deploy the front airbags because they cannot recognize oblique collision characteristics, weakening the overall protection effect.

[0006] Fourth, target type and dynamic characteristics are not fully incorporated into collision prediction models. Different types of targets (such as pedestrians, two-wheeled vehicles, passenger cars, and trucks) exhibit significantly different kinematic and dynamic responses in collisions. For example, in pedestrian collisions, the energy-absorbing design of the hood area needs to be emphasized; in two-wheeled vehicle collisions, the cyclist may impact the A-pillar area due to inertia; in truck rear-end collision scenarios, the vehicle itself risks going under the truck. Existing systems largely limit the use of target classification information to adjusting the warning level of Forward Collision Warning (FCW), failing to deeply integrate it into the collision type prediction model, thus limiting prediction accuracy.

[0007] Fifth, there is a lack of prediction confidence assessment and graded response mechanisms. Existing systems typically do not provide quantitative information on the confidence level of their predictions when outputting collision warnings or triggering safety measures. When sensor information is incomplete, environmental conditions are harsh, or the target's motion state is ambiguous, aggressive triggering strategies can easily lead to system malfunctions, while completely abandoning intervention may miss valuable opportunities for occupant protection. The lack of a confidence assessment mechanism prevents the system from implementing graded responses based on information quality, limiting the intelligence level and user acceptance of pre-collision systems.

[0008] Sixth, there is a lack of an effective mechanism for verifying the consistency of sensor data. Existing pre-collision systems assume that all sensor data is reliable input and have not established a cross-sensor data consistency verification mechanism. When a sensor malfunctions or is maliciously interfered with, the system may make incorrect decisions based on erroneous perception data, leading to safety hazards.

[0009] In summary, there is an urgent need in this field for a technical solution that can comprehensively utilize information from multiple sensors, accurately predict the type of collision that is about to occur, and optimize the control strategy of the safety restraint system accordingly, so as to overcome the shortcomings of the existing technology and improve the intelligence level and occupant protection effect of the pre-collision system. Summary of the Invention

[0010] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a collision type identification method, system, and device based on collision point prediction.

[0011] According to one aspect of this application, a collision type identification method based on collision point prediction is provided, comprising: acquiring multi-source perception data of a target vehicle; wherein the multi-source perception data includes radar data and video data; performing spatiotemporal synchronization on the multi-source perception data to obtain multi-source synchronized perception data; performing position compensation on the multi-source synchronized perception data to obtain multi-source compensated perception data; performing weighted fusion on the multi-source compensated perception data to obtain a target feature vector; comparing the radial velocity projection of the millimeter-wave radar with the velocity estimated by the optical flow field of the visual sensor, and calculating a normalized consistency score; when the normalized consistency score is low... When a preset threshold is reached, the system downgrades to basic safety mode and issues a sensor verification fault code; it acquires the state data of the target vehicle, including vehicle speed and yaw rate; based on the target feature vector and the state data, it predicts the collision point information of the target vehicle; based on the collision point information of the target vehicle, it determines the collision type and corresponding confidence level of the target vehicle; wherein the confidence level is dynamically weighted based on the normalized mutual information of the prediction results of each decision tree of the random forest classifier and the collision point prediction covariance matrix; based on the collision type and corresponding confidence level of the target vehicle, it generates a safety control strategy.

[0012] In one embodiment, the step of extracting feature data from the multi-source sensing data to obtain a target feature vector includes: performing spatiotemporal synchronization on the multi-source sensing data to obtain multi-source synchronized sensing data; performing position compensation on the multi-source synchronized sensing data to obtain multi-source compensated sensing data; and performing weighted fusion on the multi-source compensated sensing data to obtain the target feature vector.

[0013] In one embodiment, the step of spatiotemporally synchronizing the multi-source sensing data to obtain multi-source synchronized sensing data includes: performing linear interpolation based on the sampling frequency of the multi-source sensing data to obtain the multi-source synchronized sensing data.

[0014] In one embodiment, the step of performing position compensation on the multi-source synchronous sensing data to obtain multi-source compensated sensing data includes: performing position compensation on the multi-source synchronous sensing data based on the time difference between the timestamp of each point in the multi-source synchronous sensing data and the center time of the sampling frame to obtain the multi-source compensated sensing data.

[0015] In one embodiment, predicting the collision point information of the target vehicle based on the target feature vector and the state data includes: predicting the predicted motion trajectory of the target vehicle within a set time period in the future based on the state data; and predicting the collision point information of the target vehicle based on the predicted motion trajectory and the target feature vector.

[0016] In one embodiment, predicting the collision point information of the target vehicle based on the predicted motion trajectory and the target feature vector includes: calculating the probability density function value and the corresponding covariance matrix of each collision point of the target vehicle based on the position information and corresponding weight values ​​of all collision points of the target vehicle.

[0017] In one embodiment, determining the collision type and corresponding confidence level of the target vehicle based on the collision point information of the target vehicle includes: determining the collision type of the target vehicle based on the collision point information; calculating the confidence level of the collision type based on the target feature vector, the state data, and the collision point information; wherein the confidence level is dynamically weighted according to the following formula: ; in, The maximum posterior probability output by the random forest classifier is NMI, which is the normalized mutual information between the predictions of each decision tree in the random forest. NMI = MI / MI stands for Mutual Information. For the information entropy of the integrated output, NMI∈[0,1], For adaptive weighting, when the trace of the predicted covariance matrix of the collision point exceeds a threshold... Take the first value, otherwise Take the second value.

[0018] In one embodiment, generating a safety control strategy based on the collision type and corresponding confidence level of the target vehicle includes: generating the airbag inflation volume and inflation rate, and seatbelt pretension force of the target vehicle based on the collision type, corresponding confidence level, and collision angle; wherein, the actual control parameters are continuously adjusted according to the confidence level: pretension force = , This refers to the seat belt pretension force under standard operating conditions. For the confidence level of the collision type, This represents the minimum safety factor.

[0019] According to another aspect of this application, a collision type recognition system based on collision point prediction is provided, comprising: a perception data acquisition module for acquiring multi-source perception data of a target vehicle; wherein the multi-source perception data includes radar data and video data; a feature vector extraction module for spatiotemporally synchronizing the multi-source perception data to obtain multi-source synchronized perception data; performing position compensation on the multi-source synchronized perception data to obtain multi-source compensated perception data; and performing weighted fusion on the multi-source compensated perception data to obtain a target feature vector; and a sensor verification module for comparing the radial velocity projection of a millimeter-wave radar with the velocity estimated by the optical flow field of a visual sensor, and calculating a normalized consistency score; when the normalized consistency score is lower than a preset threshold... When the vehicle is in a state of emergency, it is downgraded to a basic safety mode and a sensor verification fault code is issued. A status data acquisition module is used to acquire the status data of the target vehicle, including vehicle speed and yaw rate. A collision information prediction module is used to predict the collision point information of the target vehicle based on the target feature vector and the status data. A collision type determination module is used to determine the collision type and corresponding confidence level of the target vehicle based on the collision point information, wherein the confidence level is dynamically weighted based on the normalized mutual information of the prediction results of each decision tree in the random forest classifier and the collision point prediction covariance matrix. A control strategy generation module is used to generate a safety control strategy based on the collision type and corresponding confidence level of the target vehicle.

[0020] According to another aspect of this application, an electronic device is provided, comprising: a processor; a memory for storing processor-executable instructions; the processor being configured to perform any of the methods described above.

[0021] This application provides a collision type identification method, system, and device based on collision point prediction. The method involves acquiring multi-source perception data of a target vehicle, including radar data and video data; extracting feature data from the multi-source perception data to obtain a target feature vector; acquiring the target vehicle's state data, including vehicle speed and yaw rate; predicting the target vehicle's collision point information based on the target feature vector and state data; and determining the target vehicle's collision type and corresponding confidence level based on the collision point information. The confidence level is calculated by dynamically weighting the normalized mutual information of the prediction results from each decision tree of a random forest classifier with the collision point prediction covariance matrix. Simultaneously, the method utilizes the radial velocity from millimeter-wave radar and the velocity of the visual optical flow field. Sensor consistency is verified through cross-comparison, and the system is downgraded to basic safety mode when the consistency score is below a threshold. A safety control strategy is generated based on the collision type and corresponding confidence level of the target vehicle. A target feature vector is constructed by fusing multi-source perception data to achieve high-precision perception and predict the collision point information of the target vehicle. The collision type is determined based on the collision point information, and a confidence level mechanism is introduced to determine the safety control strategy to improve the accuracy of recognition. The system dynamically generates a safety control strategy based on the recognition results. By introducing a weighted confidence level mechanism that combines a random forest classifier with a collision point uncertainty metric, the system can achieve graded response when sensor information is partially missing, thereby improving the robustness of decision-making under complex conditions and the adaptability to different collision scenarios. Attached Figure Description

[0022] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0023] Figure 1 This is a flowchart illustrating a collision type identification method based on collision point prediction provided in an exemplary embodiment of this application.

[0024] Figure 2 This is a schematic diagram of the structure of a collision type recognition system based on collision point prediction provided in an exemplary embodiment of this application.

[0025] Figure 3 This is a structural diagram of an electronic device provided in an exemplary embodiment of this application. Detailed Implementation

[0026] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0027] Figure 1 This is a flowchart illustrating a collision type identification method based on collision point prediction provided in an exemplary embodiment of this application. Figure 1 As shown, the collision type identification method based on collision point prediction includes the following steps: Step 110: Acquire multi-source perception data of the target vehicle.

[0028] The multi-source sensing data includes radar data and video data. This application uses millimeter-wave radar, lidar, and visual cameras to collect raw target sensing data.

[0029] Step 120: Perform spatiotemporal synchronization on the multi-source sensing data to obtain multi-source synchronized sensing data; perform position compensation on the multi-source synchronized sensing data to obtain multi-source compensated sensing data; perform weighted fusion on the multi-source compensated sensing data to obtain the target feature vector.

[0030] This application generates a target fusion feature vector by performing spatiotemporal synchronization, target association, and feature-level fusion on multi-sensor data. F target Spatiotemporal synchronization aligns data from different sampling frequencies to the same time reference through interpolation; position compensation corrects motion distortion; weighted fusion assigns weights based on the confidence levels of different sensors, ultimately yielding a fused feature vector.

[0031] Specifically, this application obtains multi-source synchronized sensing data by performing spatiotemporal synchronization on multi-source sensing data, and then obtains multi-source compensated sensing data by performing position compensation on the multi-source synchronized sensing data. Finally, the multi-source compensated sensing data is weighted and fused to obtain a target feature vector. The target feature vector... F target This includes: the target vehicle's geometric features (width W, height H, length L) and category features. C type (Cars / SUVs / trucks / pedestrians / two-wheeled vehicles, etc., represented using one-hot encoding), attitude features (target facing angle) Ψ target Target yaw rate ω target ), motion characteristics (relative distance) d rel Relative velocity v rel Relative acceleration arel Collision Time TTC = d rel / | v rel |, when v rel <0 is valid), trajectory features (historical location sequence), etc.

[0032] Step 130: Compare the radial velocity projection from the millimeter-wave radar with the velocity estimated by the optical flow field from the visual sensor, and calculate the normalized consistency score.

[0033] Before implementing active safety intervention strategies, a consistency check is performed on the sensors: the radial velocity projection from the millimeter-wave radar is compared with the velocity estimated by the optical flow field of the visual sensor, and a normalized consistency score is calculated. When the normalized consistency score is lower than a preset threshold multiple times consecutively, the system is downgraded to the basic safety mode (only triggering passive measures such as hazard lights and unlocking doors), and a sensor verification fault code is issued, terminating the subsequent collision prediction process. If the verification passes, the subsequent step 140 is executed. This verification runs as a periodic background task; a single verification failure does not immediately terminate the process, but only marks the sensor status to distinguish between transient disturbances and persistent faults.

[0034] Step 140: Obtain the status data of the target vehicle.

[0035] The status data includes vehicle speed and yaw rate. This application collects the vehicle speed... V ego yaw rate ω ego Longitudinal acceleration a x lateral acceleration a y Steering angle δ Brake master cylinder pressure P brake Isokinetic information.

[0036] Step 150: Based on the target feature vector and state data, predict the collision point information of the target vehicle.

[0037] This application uses an interactive multi-model (IMM) approach to predict the future trajectory of the target based on the target characteristics and the vehicle's state, and solves for the coordinates of the contact point. C(x,y) and collision angle θ impact .

[0038] Step 160: Based on the collision point information of the target vehicle, determine the collision type and corresponding confidence level of the target vehicle.

[0039] The confidence score is dynamically weighted based on the mutual information of the prediction results of each decision tree in the random forest classifier and the covariance matrix of the collision point prediction. The specific calculation method is detailed in the following example.

[0040] Step 170: Generate a safety control strategy based on the collision type and corresponding confidence level of the target vehicle.

[0041] This application adjusts the unit according to L and P(L) The triggering parameters of actuators such as airbags and seat belts are dynamically adjusted, and the actuator unit ultimately performs the corresponding actions.

[0042] This application provides a collision type identification method based on collision point prediction, which acquires multi-source perception data of the target vehicle, including radar data and video data; extracts feature data from the multi-source perception data to obtain a target feature vector; acquires the target vehicle's state data, including vehicle speed and yaw rate; predicts the target vehicle's collision point information based on the target feature vector and state data; and determines the target vehicle's collision type and corresponding confidence level based on the collision point information. The confidence level is dynamically weighted based on the normalized mutual information of the prediction results of each decision tree in a random forest classifier and the collision point prediction covariance matrix. Simultaneously, it uses the cross ratio of the millimeter-wave radar radial velocity and the visual optical flow field velocity. Sensor consistency verification is performed, and the system is downgraded to basic safety mode when the consistency score is below a threshold. A safety control strategy is generated based on the collision type and corresponding confidence level of the target vehicle. A target feature vector is constructed by fusing multi-source perception data to achieve high-precision perception and predict the collision point information of the target vehicle. The collision type is determined based on the collision point information, and a confidence level mechanism is introduced to determine the safety control strategy to improve the accuracy of recognition. The system dynamically generates a safety control strategy based on the recognition results. By introducing a weighted confidence level mechanism that combines a random forest classifier with a collision point uncertainty metric, the system can achieve graded response when sensor information is partially missing, thereby improving the robustness of decision-making under complex conditions and the adaptability to different collision scenarios.

[0043] In one embodiment, step 120 can be implemented by performing linear interpolation based on the sampling frequency of the multi-source sensing data to obtain multi-source synchronous sensing data.

[0044] To achieve accurate fusion of multi-source data, this application first performs spatiotemporal synchronization of the multi-source sensing data. Specifically, the sensor fusion unit of this application performs fusion at a fixed period. T s Data is acquired every 50ms, and data from different sampling frequencies are aligned to the same time base using interpolation. For example, the sampling frequency of millimeter-wave radar is... f r=20Hz, the sampling frequency of the lidar is f r =10Hz, the camera's sampling frequency is f c =30Hz, then using the highest frequency of 30Hz as the reference, linear interpolation is performed on the other sensor data: for any time... t， sensor s The interpolated data are: ; in, For sensors s At any moment t interpolated data, For sensors s At any moment The sampling data, For sensors s At any moment The sampling data.

[0045] In one embodiment, step 120 can be implemented by performing position compensation on the multi-source synchronous sensing data based on the time difference between the timestamp of each point in the multi-source synchronous sensing data and the center time of the sampling frame, to obtain multi-source compensated sensing data.

[0046] To address motion distortion caused by sampling time discrepancies between different sensors, this application employs an adaptive motion compensation factor. Specifically, let the vehicle speed between two adjacent frames be... v ego The relative speed between the target object and the target vehicle is v rel Then, for each point in the lidar point cloud, use its timestamp... With frame center time The difference Perform position compensation: ; in, The position after compensation. The position before compensation. a rel The relative acceleration is estimated recursively using a Kalman filter. This compensation method effectively eliminates the point cloud "tailing" effect caused by target motion during the scanning cycle in rotating scanning lidar, improving the spatiotemporal consistency of the fused features.

[0047] In one embodiment, step 130 can be implemented as follows: using a pre-calibrated radar-camera extrinsic and intrinsic parameter matrix, each target point detected by the millimeter-wave radar is projected onto the image pixel coordinate system of the visual sensor to obtain the corresponding pixel coordinates; then, at the pixel coordinates, an optical flow algorithm (such as the Lucas-Kanade algorithm) is used to calculate the optical flow vector of that point based on two consecutive frames of images. This serves as the two-dimensional velocity vector estimated by the visual sensor; next, the radial velocity v measured by the radar is... radar By combining the target point's three-dimensional position information, a three-dimensional velocity vector is reconstructed and projected onto the image plane to obtain the radar-estimated two-dimensional velocity vector. Calculate the Euclidean distance between the two as the velocity residual: Finally, the velocity residuals of all N valid targets in the current frame are negatively exponentially normalized and averaged to obtain the normalized consistency score: ; in, These are preset scaling parameters used to control the degree of influence of residuals on scores; The value range is [0,1], and the closer the value is to 1, the more consistent the speed observations of the radar and the visual sensor are.

[0048] like S cons If the value is less than 0.3, the sensor data is deemed unreliable, the system switches to basic safety mode (only triggering passive measures such as hazard lights and unlocking doors) and issues a fault code, and no longer performs subsequent collision prediction and active intervention.

[0049] In one embodiment, step 150 can be implemented as follows: based on state data, predict the predicted trajectory of the target vehicle within a set time period in the future; based on the predicted trajectory and the target feature vector, predict the collision point information of the target vehicle.

[0050] The target vehicle's state perception module collects dynamic information such as vehicle speed and yaw rate in real time, and uses a constant turning rate and speed model to predict the target vehicle's trajectory within the next 0.5 to 1.5 seconds. The prediction formula for the trajectory within the set future time period is:

[0051] ; in, They are respectively The longitudinal position, lateral position, heading angle, vehicle speed, and yaw rate at any given moment. They are respectively The longitudinal position, lateral position, heading angle, vehicle speed, and yaw rate at any given moment.

[0052] In one embodiment, step 150 can be implemented by: calculating the probability density function value and the corresponding covariance matrix of each collision point of the target vehicle based on the location information and corresponding weight values ​​of all collision points of the target vehicle.

[0053] When the TTC of any target is less than the preset activation threshold T active At a time of 1.5 seconds (e.g.), the collision point prediction module is activated. Target trajectory prediction employs an interactive multi-model approach, fusing three motion models: constant velocity, constant acceleration, and constant turning rate. Simultaneously, the target's reasonable outline is simplified to a rectangle, and the separating axis theorem is used for collision angle detection. The formula for calculating the collision angle is: ; in, For the collision angle, For the target vehicle speed vector, This is a unit vector representing the direction the target vehicle is facing (front direction), with 0° indicating a head-on collision and 180° indicating a rear-end collision. This approach essentially categorizes collision directions as frontal, side, or rear-end, facilitating safety strategy matching. The specific value is determined by the location of the contact point on the target vehicle at the time of the collision: for a frontal collision (such as a head-on impact), the contact point normal points towards the front of the vehicle; for a rear-end collision (such as a rear-end collision), the contact point normal points towards the rear of the vehicle; for a left-side collision, the contact point normal points to the left; and for a right-side collision, the contact point normal points to the right.

[0054] To quantify prediction uncertainty, particle filtering is used to estimate the probability density of the contact point coordinates. Let the particle set... ,in, For the first i The position of each particle For the first i The weight of each particle, Let be the number of particles in the particle cluster. Then the collision point... C The distribution is given by the weighted kernel density estimate: ; in, The point of collision C The probability density function value, For kernel function, For the first i The hypothetical coordinates of the collision point location represented by each particle are used. The final output contact point coordinates are taken as the mean of the distribution, and the covariance matrix is ​​output as a measure of the uncertainty of the collision point prediction. This measure will be combined with the classifier output in the subsequent confidence evaluation.

[0055] In one embodiment, step 160 can be implemented as follows: determining the collision type of the target vehicle based on the collision point information; and calculating the confidence level of the collision type based on the target feature vector, state data, and collision point information.

[0056] This application establishes 12 label libraries (L01~L12 as shown in Table 1 below), and the classification rules are based on geometric features and motion parameters, using a combination of rule-based decision trees and random forests. The random forest is input feature vectors. X= [ x c ,y c ,θ impact ,r y , d rel ,v rel ,TTC,onehot(C type ) ],in, x c 、y c The collision point coordinates (i.e., the collision point coordinates used to determine the collision type in Table 1) x c 、y c ), r y The lateral deviation is expressed as the posterior probability P(L|X).

[0057] Confidence P conf The calculation formula is: ; in, The maximum a posteriori probability of the random forest output; NMI is the normalized mutual information, NMI = MI / H ( ), where MI stands for mutual information, and H( (for integrated output) Information entropy, NMI∈[0,1], It is obtained by averaging the predictions of all decision trees. , This means taking the average of all decision trees. For adaptive weights, when the uncertainty of collision point prediction tr(ΣC) > τ cov hour, =0.4 (reduces dependence on classifier output, focuses more on model consistency), otherwise =0.7, where ΣC is the covariance matrix of the collision point C, tr(ΣC) represents the trace of the covariance matrix (the sum of the diagonal elements), τ cov To set a threshold.

[0058] when P conf <γ ( γ When the threshold is 0.7 and MI>0.5, it is judged as "high uncertainty - unreliable type". The system not only switches to a conservative strategy (the threshold is set to the lowest value, the front and side airbag circuits are preset, and the seat belt is 1.2 times the standard), but also actively reduces the trigger threshold of the safety restraint system and extends the sensor refresh cycle (to 10ms) to obtain more information.

[0059] Table 1. Definitions and criteria for collision types

[0060] Note: ε=0.1m is the position tolerance, derived from the typical positioning accuracy of sensor fusion. For the pole collision (L11), excluding the target width W... t In addition to being ≤0.5m in height and being a static obstacle, the target must also have an aspect ratio <0.3 (i.e., exhibit a distinct columnar shape). This feature mainly relies on the shape recognition results of LiDAR point cloud data to reduce the visual false detection rate.

[0061] In one embodiment, step 170 can be implemented by generating the airbag inflation volume and inflation rate, and seat belt pretension force of the target vehicle based on the collision type and corresponding confidence level and collision angle of the target vehicle.

[0062] Among them, airbag trigger threshold Th airbag =k type ·Th std , Th std This is the trigger value for a specific collision scenario. Th std =40g, k type The proportionality coefficient was obtained through multibody dynamics simulation optimization, and the specific results are shown in the table below. A proportionality coefficient less than 1 indicates a lower trigger threshold, allowing the system to intervene earlier. For scenarios with low overlap rates and side collisions that are prone to causing large displacements of occupants, appropriately earlier triggering is beneficial for gaining protection time.

[0063] Table 2 Proportion Coefficient Table

[0064] Airbag deployment method: For oblique angle collisions, this application employs an inflation volume distribution strategy based on the collision angle, wherein, V std The formula for calculating the airbag inflation volume under standard operating conditions is: V front =V std ·cos(θ impact ), V side =V std sin(θ) impact ) ; in, V front For the inflation volume of the front airbag, V side Inflation volume of the side airbags.

[0065] For oblique collisions (L06 / L07), proportional inflation rate control is also proposed. The frontal and lateral inflation rates satisfy total volume conservation. ， ； in, For the inflation rate of the front airbag, This refers to the rated inflation rate of the airbag under standard operating conditions. The side airbag inflation rate is set at a slower rate, while the side airbags inflate rapidly to their predetermined capacity. This prevents the front airbags from deploying prematurely and hindering occupant lateral movement in an angled collision. This rate distribution strategy has been validated by MADYMO simulations and effectively reduces the chest viscosity index under typical angled collision conditions (e.g., 64 km / h, 25° angle).

[0066] Seat belt pretension F pretension =β type ·F std ，β type The seatbelt pretension factor (F is shown in the table below) is... std This refers to the seat belt pretension force under standard operating conditions. F std =2000N. The actual preload is also continuously adjusted according to the confidence level: , α=0.6.

[0067] Table 3 Seat belt pretension coefficient table

[0068] Figure 2 This is a schematic diagram of the structure of a collision type recognition system based on collision point prediction provided in an exemplary embodiment of this application. Figure 2 As shown, the collision type recognition system 20 based on collision point prediction includes: a perception data acquisition module 21, used to acquire multi-source perception data of the target vehicle; wherein, the multi-source perception data includes radar data and video data; a feature vector extraction module 22, used to perform spatiotemporal synchronization of the multi-source perception data to obtain multi-source synchronized perception data; perform position compensation on the multi-source synchronized perception data to obtain multi-source compensated perception data; and perform weighted fusion of the multi-source compensated perception data to obtain a target feature vector; and a sensor verification module 23, used to compare the radial velocity projection of the millimeter-wave radar with the velocity estimated by the optical flow field of the visual sensor, and calculate a normalized consistency score; when the normalized consistency score is lower than a preset threshold... When the value is low, the system downgrades to the basic safety mode and issues a sensor verification fault code; the status data acquisition module 24 is used to acquire the status data of the target vehicle, including vehicle speed and yaw rate; the collision information prediction module 25 is used to predict the collision point information of the target vehicle based on the target feature vector and status data; the collision type determination module 26 is used to determine the collision type and corresponding confidence level of the target vehicle based on the collision point information of the target vehicle, wherein the confidence level is dynamically weighted based on the normalized mutual information of the prediction results of each decision tree of the random forest classifier and the collision point prediction covariance matrix; the control strategy generation module 27 is used to generate a safety control strategy based on the collision type and corresponding confidence level of the target vehicle.

[0069] This application provides a collision type identification system based on collision point prediction. A perception data acquisition module 21 acquires multi-source perception data of the target vehicle, including radar data and video data. A feature vector extraction module 22 extracts feature data from the multi-source perception data to obtain a target feature vector. A sensor verification module 23 compares the radial velocity projection from the millimeter-wave radar with the velocity estimated by the optical flow field of the visual sensor, calculating a normalized consistency score. When the normalized consistency score is lower than a preset threshold, the system downgrades to a basic safety mode and issues a sensor verification fault code. A state data acquisition module 24 acquires state data of the target vehicle, including vehicle speed and yaw rate. A collision information prediction module 25 predicts the collision point information of the target vehicle based on the target feature vector and state data. A collision type determination module 26 determines the collision type based on the target vehicle's collision point information. The system identifies the collision type and corresponding confidence level of the target vehicle based on the collision point information. The confidence level is dynamically weighted based on the mutual information of the prediction results of each decision tree in the random forest classifier and the covariance matrix of the collision point prediction. The control strategy generation module 27 generates a safety control strategy based on the collision type and corresponding confidence level of the target vehicle. By fusing multi-source sensing data to construct a target feature vector, high-precision sensing is achieved, and the collision point information of the target vehicle is predicted. Based on the collision point information, the collision type is determined and a confidence level mechanism is introduced to determine the safety control strategy, thereby improving the accuracy of recognition. The system dynamically generates a safety control strategy based on the recognition results. By introducing a weighted confidence level mechanism of random forest classifier and collision point uncertainty measurement, the system can achieve hierarchical response when sensor information is partially missing, thereby improving the robustness of decision-making under complex working conditions and the adaptability to different collision scenarios.

[0070] In one embodiment, the feature vector extraction module 22 can be further configured to: perform linear interpolation based on the sampling frequency of the multi-source sensing data to obtain multi-source synchronous sensing data.

[0071] In one embodiment, the feature vector extraction module 22 can be further configured to: perform position compensation on the multi-source synchronous sensing data based on the time difference between the timestamp of each point in the multi-source synchronous sensing data and the center time of the sampling frame, to obtain multi-source compensated sensing data.

[0072] In one embodiment, the collision information prediction module 25 can be further configured to: predict the predicted motion trajectory of the target vehicle within a set time period based on state data; and predict the collision point information of the target vehicle based on the predicted motion trajectory and the target feature vector.

[0073] In one embodiment, the collision information prediction module 25 can be further configured to: calculate the probability density function value and the corresponding covariance matrix of each collision point of the target vehicle based on the position information and corresponding weight values ​​of all collision points of the target vehicle.

[0074] In one embodiment, the collision type determination module 26 can be further configured to: determine the collision type of the target vehicle based on the collision point information; and calculate the confidence level of the collision type based on the target feature vector, state data, and collision point information, wherein the confidence level is dynamically weighted according to the following formula: ,in, The maximum posterior probability output by the random forest classifier is NMI, which is the normalized mutual information. NMI = MI / H ), where MI stands for mutual information, and H( (for integrated output) Information entropy, NMI∈[0,1], It is obtained by averaging the predictions of all decision trees. , This means taking the average of all decision trees. For adaptive weights, when the uncertainty of collision point prediction tr(ΣC) > τ cov hour, =0.4 (reduces dependence on classifier output, focuses more on model consistency), otherwise =0.7, where ΣC is the covariance matrix of the collision point C, tr(ΣC) represents the trace of the covariance matrix (the sum of the diagonal elements), τ cov To set a threshold.

[0075] In one embodiment, the control strategy generation module 27 can be further configured to generate the airbag inflation volume and inflation rate, and seat belt pretension force of the target vehicle based on the collision type and corresponding confidence level and collision angle of the target vehicle.

[0076] Below, for reference Figure 3 This application describes an electronic device according to embodiments thereof. The electronic device may be either or both of a first device and a second device, or a standalone device independent of them, which may communicate with the first device and the second device to receive acquired input signals from them.

[0077] Figure 3 A block diagram of an electronic device according to an embodiment of this application is illustrated.

[0078] like Figure 3 As shown, the electronic device 10 includes one or more processors 11 and memory 12.

[0079] The processor 11 may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.

[0080] The memory 12 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 11 may execute the program instructions to implement the methods of the various embodiments of this application described above and / or other desired functions. Various contents such as input signals, signal components, and noise components may also be stored in the computer-readable storage medium.

[0081] In one example, the electronic device 10 may also include an input device 13 and an output device 14, which are interconnected via a bus system and / or other forms of connection mechanism (not shown).

[0082] When the electronic device is a standalone device, the input device 13 can be a communication network connector for receiving the collected input signals from the first device and the second device.

[0083] In addition, the input device 13 may also include, for example, a keyboard, a mouse, etc.

[0084] The output device 14 can output various information to the outside, including determined distance information, direction information, etc. The output device 14 may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0085] Of course, for the sake of simplicity, Figure 3 Only some of the components of the electronic device 10 relevant to this application are shown in this illustration; components such as buses, input / output interfaces, etc., are omitted. In addition, the electronic device 10 may include any other suitable components depending on the specific application.

[0086] In addition to the methods and apparatus described above, embodiments of this application may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform the steps in the methods according to various embodiments of this application described in the "Exemplary Methods" section above.

[0087] The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of this application. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0088] Furthermore, embodiments of this application may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps in the methods according to various embodiments of this application described in the "Exemplary Methods" section above.

[0089] The computer-readable storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, 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 device, magnetic storage device, or any suitable combination thereof.

[0090] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.

[0091] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0092] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0093] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0094] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A collision type identification method based on collision point prediction, characterized in that, include: Acquire multi-source perception data of the target vehicle; wherein, the multi-source perception data includes radar data and video data; The multi-source sensing data is spatiotemporally synchronized to obtain multi-source synchronized sensing data; the multi-source synchronized sensing data is positionally compensated to obtain multi-source compensated sensing data; the multi-source compensated sensing data is weighted and fused to obtain the target feature vector. The radial velocity projection of the millimeter-wave radar is compared with the velocity estimated by the optical flow field of the visual sensor to calculate a normalized consistency score; when the normalized consistency score is lower than a preset threshold, the system is downgraded to the basic safety mode and a sensor verification fault code is issued. Acquire the state data of the target vehicle; wherein the state data includes vehicle speed and yaw rate; Based on the target feature vector and the state data, predict the collision point information of the target vehicle; Based on the collision point information of the target vehicle, the collision type and corresponding confidence level of the target vehicle are determined; wherein, the confidence level is dynamically weighted based on the normalized mutual information of the prediction results of each decision tree of the random forest classifier and the collision point prediction covariance matrix. A safety control strategy is generated based on the collision type and corresponding confidence level of the target vehicle.

2. The collision type identification method based on collision point prediction according to claim 1, characterized in that, The process of spatiotemporally synchronizing the multi-source sensing data to obtain multi-source synchronized sensing data includes: Based on the sampling frequency of the multi-source sensing data, linear interpolation is performed to obtain the multi-source synchronous sensing data.

3. The collision type identification method based on collision point prediction according to claim 1, characterized in that, The step of performing position compensation on the multi-source synchronous sensing data to obtain multi-source compensated sensing data includes: Based on the time difference between the timestamp of each point in the multi-source synchronous sensing data and the center time of the sampling frame, position compensation is performed on the multi-source synchronous sensing data to obtain the multi-source compensated sensing data.

4. The collision type identification method based on collision point prediction according to claim 1, characterized in that, The step of predicting the collision point information of the target vehicle based on the target feature vector and the state data includes: Based on the state data, predict the target vehicle's trajectory within a set future time period; Based on the predicted motion trajectory and the target feature vector, the collision point information of the target vehicle is predicted.

5. The collision type identification method based on collision point prediction according to claim 4, characterized in that, The step of predicting the collision point information of the target vehicle based on the predicted motion trajectory and the target feature vector includes: Based on the location information and corresponding weight values ​​of all collision points of the target vehicle, the probability density function value and the corresponding covariance matrix of each collision point of the target vehicle are calculated.

6. The collision type identification method based on collision point prediction according to claim 1, characterized in that, The step of determining the collision type and corresponding confidence level of the target vehicle based on the collision point information of the target vehicle includes: Based on the collision point information, the collision type of the target vehicle is determined; Based on the target feature vector, the state data, and the collision point information, the confidence level of the collision type is calculated; wherein, the confidence level is dynamically weighted according to the following formula: ； in, The maximum posterior probability output by the random forest classifier is NMI, which is the normalized mutual information between the predictions of each decision tree in the random forest. NMI = MI / MI stands for Mutual Information. For the information entropy of the integrated output, NMI∈[0,1], For adaptive weighting, when the trace of the predicted covariance matrix of the collision point exceeds a threshold... Take the first value, otherwise Take the second value.

7. The collision type identification method based on collision point prediction according to claim 1, characterized in that, The generation of a safety control strategy based on the collision type and corresponding confidence level of the target vehicle includes: Based on the collision type and corresponding confidence level and collision angle of the target vehicle, the airbag inflation volume and inflation rate, and seat belt pretension force of the target vehicle are generated; wherein, the seat belt pretension force = , This refers to the seat belt pretension force under standard operating conditions. For the confidence level of the collision type, This represents the minimum safety factor.

8. A collision type recognition system based on collision point prediction, characterized in that, include: The perception data acquisition module is used to acquire multi-source perception data of the target vehicle; wherein, the multi-source perception data includes radar data and video data; The feature vector extraction module is used to perform spatiotemporal synchronization on the multi-source sensing data to obtain multi-source synchronized sensing data; perform position compensation on the multi-source synchronized sensing data to obtain multi-source compensated sensing data; and perform weighted fusion on the multi-source compensated sensing data to obtain the target feature vector. The sensor verification module is used to compare the radial velocity projection of the millimeter-wave radar with the velocity estimated by the optical flow field of the visual sensor and calculate the normalized consistency score; when the normalized consistency score is lower than a preset threshold, it is downgraded to the basic safety mode and a sensor verification fault code is issued. A status data acquisition module is used to acquire the status data of the target vehicle; wherein, the status data includes vehicle speed and yaw rate; The collision information prediction module is used to predict the collision point information of the target vehicle based on the target feature vector and the state data. The collision type determination module is used to determine the collision type and corresponding confidence level of the target vehicle based on the collision point information of the target vehicle; wherein, the confidence level is dynamically weighted based on the normalized mutual information of the prediction results of each decision tree of the random forest classifier and the collision point prediction covariance matrix. The control strategy generation module is used to generate a safety control strategy based on the collision type and corresponding confidence level of the target vehicle.

9. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is used to execute the method described in any one of claims 1-7.

Images