Search

CN-122009161-A - Contact risk prediction method for vehicle chassis and vehicle

CN122009161ACN 122009161 ACN122009161 ACN 122009161ACN-122009161-A

Abstract

The application relates to a contact risk prediction method of a vehicle chassis and a vehicle, belonging to the technical field of risk prediction, comprising the following steps: obtaining vehicle driving data of a target scene, extracting three-dimensional features of a road surface in front of the vehicle in the vehicle driving data to obtain a road feature tensor, predicting the distance between a vehicle chassis and the road surface in front of the vehicle and the distance change trend step by step based on the road feature tensor, presetting vehicle structural parameters and the vehicle driving data, respectively obtaining a minimum ground clearance value and a ground clearance change trend, carrying out feature fusion with the road feature tensor to obtain a fusion feature vector, inputting the fusion feature vector into a pre-trained contact probability judging model to obtain a chassis contact risk level, generating a chassis contact early warning signal based on the chassis contact risk level, and realizing advanced and accurate quantitative evaluation of the chassis contact risk, thereby remarkably improving the risk early warning capability and the driving safety of the vehicle under the off-road scene.

Inventors

  • LI MING

Assignees

  • 长城汽车股份有限公司

Dates

Publication Date
20260512
Application Date
20260126

Claims (10)

  1. 1. A method of predicting risk of contact of a vehicle chassis, comprising: acquiring vehicle driving data of a target scene; extracting three-dimensional features of a road surface in front of a vehicle in the vehicle driving data to obtain a road feature tensor; Step-by-step prediction is carried out on the distance between a vehicle chassis and a road surface in front of the vehicle and the distance change trend based on the road characteristic tensor, preset vehicle structural parameters and the vehicle driving data, so as to respectively obtain a minimum ground clearance value and a ground clearance change trend; performing feature fusion on the road feature tensor, the minimum ground clearance value and the ground clearance change trend to obtain a fusion feature vector; And inputting the fusion feature vector into a pre-trained contact probability judging model to obtain a chassis contact risk level, and generating a chassis contact early warning signal based on the chassis contact risk level.
  2. 2. The method for predicting the contact risk of a vehicle chassis according to claim 1, wherein extracting three-dimensional features of a road surface in front of a vehicle in the vehicle driving data to obtain a road feature tensor comprises: based on time sequence image data in the vehicle driving data, extracting front view angle image features through a space perception network; performing space coordinate mapping on the front view image characteristics to obtain dense space characteristics; And carrying out three-dimensional curvature decoding on the dense space features, extracting the three-dimensional features of the road surface in front of the vehicle in the longitudinal direction and the transverse direction, and obtaining the road feature tensor.
  3. 3. The method for predicting the contact risk of the vehicle chassis according to claim 2, wherein the minimum ground clearance value is obtained by performing nonlinear regression mapping processing by a multi-layer perceptron based on the road feature tensor, the preset vehicle structural parameter and vehicle speed data in the vehicle driving data.
  4. 4. The contact risk prediction method of a vehicle chassis according to claim 3, wherein a time series sequence is constituted based on the vehicle speed data, vehicle posture time series data in the vehicle driving data, the road characteristic tensor, and the minimum ground clearance value; Encoding the time sequence dependency relationship in the time sequence through a long-short-term memory network, and extracting time sequence characteristics representing the dynamic response rule of the vehicle chassis; And carrying out regression prediction processing on the time sequence characteristics to obtain the change trend of the ground clearance in a future preset time period.
  5. 5. The method for predicting the contact risk of the vehicle chassis according to claim 1, wherein the feature fusion is performed on the road feature tensor, the minimum ground clearance value and the trend of the ground clearance change, so as to obtain a fusion feature vector, including: performing principal component analysis on the road feature tensor, and extracting a principal direction terrain change index to obtain a first feature sub-vector; Extracting a chassis static response characteristic in the minimum ground clearance value, and splicing the chassis static response characteristic with the first characteristic subvector to obtain a structural terrain coupling characteristic; Extracting statistical features in the change trend of the ground clearance to obtain a second feature sub-vector; and splicing the structural terrain coupling feature and the second feature sub-vector, and carrying out self-adaptive fusion through an attention weighting mechanism to obtain the fusion feature vector.
  6. 6. The method of claim 4, wherein the physical contact signal is obtained by a contact sensor on the vehicle chassis; Generating three-dimensional point cloud data of geometric forms of road surfaces in front of vehicles through three-dimensional reconstruction based on time sequence image data in the vehicle driving data; And labeling a physical bottom scraping event based on the physical contact signal, the three-dimensional point cloud data and a preset vehicle chassis design model, and generating a bottom scraping grade label.
  7. 7. The method for predicting the contact risk of the vehicle chassis according to claim 6, wherein a plurality of classifiers are integrated through an integration algorithm based on the fusion feature vector and the bottom scraping grade label, and the output of each classifier is fused through a voting mechanism to construct a contact probability discrimination model; And performing supervision training on the contact probability judgment model through the fusion feature vector and the bottom scraping grade label to obtain a pre-trained contact probability judgment model.
  8. 8. The method for predicting the risk of contact of a vehicle chassis according to claim 7, wherein inputting the fusion feature vector into a pre-trained contact probability discrimination model to obtain a chassis contact risk level comprises: Respectively outputting the probability that the fusion feature vector belongs to each category in the bottom scraping grade label through each classifier in the contact probability judging model; And carrying out weighted voting decision on the probabilities output by the classifiers to obtain the chassis contact risk level.
  9. 9. The method of claim 7, wherein the spatial perception network, the multi-layer perception machine and the long-short term memory network are each independently pre-trained; Freezing preset network parameters in the space perception network, the multi-layer perception machine and the long-period memory network; and thawing the preset network parameters layer by layer, inputting the fusion feature vector into the contact probability discrimination model, and carrying out joint training on the space perception network, the multi-layer perception machine, the long-short-period memory network and the contact probability discrimination model through a joint loss function.
  10. 10. A vehicle comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the contact risk prediction method of a vehicle chassis according to any of claims 1 to 9 when the computer program is executed.

Description

Contact risk prediction method for vehicle chassis and vehicle Technical Field The application relates to the technical field of risk prediction, in particular to a contact risk prediction method of a vehicle chassis and a vehicle. Background Along with the application of the intelligent off-road vehicle in complex terrains, the requirements of prospective and accurate prediction on the contact risk of the vehicle chassis and the ground are increasingly highlighted, and accurate prediction on the contact probability of the chassis is a key for improving the trafficability of the vehicle, guaranteeing the safety of the chassis and optimizing the driving experience, and is one of the core technical challenges for realizing the high-order off-road assistance or automatic driving function. At present, in the existing method, the road condition in front of the vehicle is mainly sensed through a vehicle-mounted sensor or combined with a high-precision map, and simple trafficability judgment is carried out based on preset static parameters such as a vehicle ground clearance or an inertia measurement unit is introduced to acquire the real-time posture and motion state of the vehicle, so that risk assessment is assisted. In the prior art, due to the fact that the two-dimensional or low-dimensional environment perception information is relied on, the three-dimensional geometric curvature of a road in the longitudinal direction and the transverse direction is difficult to reconstruct accurately, meanwhile, the coupling modeling of the dynamic response of a vehicle motion state and a chassis is also lacked, and therefore a complete prediction chain from the running speed, the path characteristics and the vehicle body posture to the chassis contact state cannot be established by the system, the chassis contact probability is difficult to evaluate in advance and quantitatively, and the accuracy and the timeliness of risk early warning are limited. Disclosure of Invention The present application solves at least one of the technical problems in the related art to a certain extent. Therefore, the application aims to provide a contact risk prediction method of a vehicle chassis and a vehicle, which are used for predicting the distance between the vehicle chassis and a road surface in front of the vehicle and the change trend of the distance by extracting the three-dimensional characteristics of the road in front of the vehicle, so as to predict the contact risk, realize the advanced and accurate quantitative evaluation of the chassis contact risk, remarkably improve the risk early warning capability and the driving safety of the vehicle in an off-road scene, and solve the problems of early warning lag and insufficient accuracy of the chassis contact risk caused by the lack of the three-dimensional road modeling capability and the dynamic response prediction mechanism in the complex terrain in the prior art. To achieve the above object, the present application provides, in a first aspect, a contact risk prediction method for a vehicle chassis, including: acquiring vehicle driving data of a target scene; extracting three-dimensional features of a road surface in front of a vehicle in the vehicle driving data to obtain a road feature tensor; Step-by-step prediction is carried out on the distance between a vehicle chassis and a road surface in front of the vehicle and the distance change trend based on the road characteristic tensor, preset vehicle structural parameters and the vehicle driving data, so as to respectively obtain a minimum ground clearance value and a ground clearance change trend; performing feature fusion on the road feature tensor, the minimum ground clearance value and the ground clearance change trend to obtain a fusion feature vector; And inputting the fusion feature vector into a pre-trained contact probability judging model to obtain a chassis contact risk level, and generating a chassis contact early warning signal based on the chassis contact risk level. In the prior art, the off-road vehicle perceives a front obstacle mainly through two-dimensional vision or radar, static trafficability judgment is carried out by combining fixed geometric parameters of the vehicle, and it is difficult to accurately model the coupling relation between the curvature of the three-dimensional road and the dynamic response of the vehicle, so that hysteresis and high misjudgment rate exist in the prediction of chassis contact risk. According to the technical scheme, vehicle driving data containing multidimensional information is firstly obtained, road feature tensors representing three-dimensional geometric features of roads are extracted from the vehicle driving data, structural modeling of complex terrains is achieved, step-by-step prediction is further carried out on chassis ground clearance and change trend of the chassis ground clearance based on the road feature tensors, vehicle structural parameters and real-time drivi