Search

CN-121989996-A - Vehicle control method, system and electronic equipment

CN121989996ACN 121989996 ACN121989996 ACN 121989996ACN-121989996-A

Abstract

The application discloses a vehicle control method, a vehicle control system and electronic equipment. The method comprises the steps of obtaining driving state information of one or more dimensions of a driver, analyzing the driving state information by using a machine learning model to obtain a driving incapacity risk index of the driver, wherein the driving incapacity risk index is used for representing the probability or degree of incapacity of the driver to safely drive, and taking over a vehicle driven by the driver under the condition that the driving incapacity risk index is larger than a preset threshold value. The application solves the technical problem that the vehicle is out of control continuously in a non-response state because the related technology cannot actively take over the control of the vehicle when the driver is suddenly disabled.

Inventors

  • LU JUN

Assignees

  • 中国第一汽车股份有限公司

Dates

Publication Date
20260508
Application Date
20260325

Claims (13)

  1. 1.A vehicle control method characterized by comprising: acquiring driving state information of one or more dimensions of a driver; analyzing the driving state information by using a machine learning model to obtain a driving incapacitation risk index of the driver, wherein the driving incapacitation risk index is used for representing the probability or degree of incapacitation of the driver for safe driving; And taking over the vehicle driven by the driver under the condition that the driving disability risk index is larger than a preset threshold value.
  2. 2. The method of claim 1, wherein the driving state information of the plurality of dimensions comprises a physiological signal, a visual behavior signal, an audio signal, and a driving operation trigger signal, and wherein the machine learning model is trained by: Acquiring sample data, wherein the sample data carries a label marked with an incapacitation state, and the sample data comprises synchronously acquired contact type physiological and electric characteristic data, non-contact type visual behavior characteristic data, audio characteristic data and driving control discrete signals; based on a unified time reference, performing time domain alignment on the contact type physiological and electric characteristic data, the non-contact type visual behavior characteristic data, the audio characteristic data and the driving control discrete signal, and performing synchronous slicing on the aligned data according to a preset time length to obtain a multi-mode synchronous original data fragment; Extracting physiological feature vectors, visual feature vectors and audio feature vectors from the multi-modal synchronous original data segment; Performing space mapping on the physiological feature vector, the visual feature vector and the audio feature vector to obtain multi-mode mapping features; constructing a driving state space-time correlation tensor according to the multi-modal mapping characteristics and the control deviation values of the driving control discrete signals corresponding to the multi-modal synchronous original data fragments; According to the change rate of each modal feature component in the driving state space-time correlation tensor, determining the feature mutation strength of the corresponding dimension of the modal feature component, and according to the feature mutation strength, calculating the contribution weight coefficient of each modal feature component in the driving state space-time correlation tensor in the corresponding time step; And taking the fusion characteristic weighted by the contribution weight coefficient as a model input, taking the label as a prediction target, and minimizing a preset loss function by using a gradient descent algorithm until parameter convergence of the machine learning model is completed.
  3. 3. The method according to claim 2, wherein calculating the contribution weight coefficients of each modal feature component in the driving state spatiotemporal association tensor at the corresponding time step comprises: acquiring acceleration characteristics representing the vibration intensity of the vehicle and visual quality characteristics representing the change of illumination environment, and acquiring driving behavior information and external environment information; according to the acceleration characteristics and the speed information of the vehicle in the current dynamic state of the vehicle, performing motion artifact correlation analysis on the physiological characteristic vector to obtain an analysis result; According to the visual quality characteristics and the illumination intensity in the external environment information, performing imaging definition evaluation on the visual characteristic vector to obtain an evaluation result; determining the credibility score of each modal feature component in the driving state space-time correlation tensor according to the analysis result and the evaluation result; According to steering angle and gear information in the driving behavior information and road topological features in the external environment information, carrying out semantic weighting processing on the physiological feature vector, the visual feature vector and the audio feature vector to obtain a scene semantic gain coefficient for reflecting the urgency degree of the current driving scene; And comprehensively correcting the characteristic mutation strength by utilizing the credibility score and the scene semantic gain coefficient to obtain corrected mutation strength, and calculating the contribution weight coefficient of each modal characteristic component in the driving state space-time correlation tensor in the corresponding time step based on the corrected mutation strength.
  4. 4. The method of claim 1, wherein after obtaining the driving disability risk indicator for the driver, the method further comprises: acquiring a domain knowledge base formed by a plurality of priori logic rules; Converting driving behavior information, external environment information and the driving disability risk index into logic facts which can be identified by answer set programming; Inputting the logic facts into an inference engine programmed based on the answer set, calling prior logic rules in the domain knowledge base through the inference engine, and carrying out logic conflict detection and consistency verification on the input logic facts; if logic conflict exists, searching an answer set meeting the prior logic rule by utilizing the reasoning engine, carrying out consistency correction on a predicted decision action corresponding to the driving incapacitation risk index according to the answer set to obtain a target execution instruction, and sending the target execution instruction to a braking system of the vehicle; And outputting a rule violation item causing the logic conflict, wherein the rule violation item is used as an interpretability basis of the driving disability risk indicator.
  5. 5. The method of claim 1, wherein taking over the vehicle driven by the driver comprises: Executing longitudinal dynamics control to reduce the running speed of the vehicle below a preset safety threshold; Planning a local target path from the current position to the nearest safe parking spot in real time based on map data and external environment information; and driving the vehicle to travel along the local target path and reducing the travel speed of the vehicle to zero to realize parking control under the condition that the travel speed is below the preset safety threshold and the external environment information meets the preset lane change condition.
  6. 6. The method of claim 5, wherein reducing the travel speed of the vehicle below a preset safety threshold comprises: Acquiring a real-time risk level corresponding to the driving incapacitation risk index; Calculating a target deceleration value and a change rate of the target deceleration value according to the real-time risk level, wherein the higher the real-time risk level is, the larger the absolute values of the corresponding target deceleration value and the change rate are; and executing a deceleration action according to the target deceleration value and the change rate by a drive-by-wire system or a drive-by-wire system until the running speed of the vehicle falls below the preset safety threshold.
  7. 7. The method of claim 1, further comprising, while taking over the vehicle driven by the driver, performing the steps of: triggering an early warning signal according to the grade of the driving incapacitation risk index, wherein the early warning signal comprises at least one of screen vision flickering reminding, audio sound reminding and seat touch vibration reminding; Broadcasting a current takeover state and a subsequent driving intention, wherein the driving intention comprises at least one of a deceleration avoidance intention, a lane changing and side leaning intention and a trace seeking and parking intention; detecting whether the driver resumes the physical take-over action of the vehicle, and dynamically adjusting the strength of the early warning signal or the take-over instruction according to the detection result.
  8. 8. The method of claim 1, wherein after taking over the vehicle driven by the driver, the method further comprises: activating an emergency call service of the vehicle to initiate a rescue request to a preset rescue center; Invoking a sensor within the vehicle to sense a current passenger number; Acquiring basic rescue information and generating an accident type label, wherein the basic rescue information comprises geographic position coordinates and a unique vehicle identification code, and the accident type label at least comprises the driving incapacitation risk index; and sending a target message to the rescue center, wherein the target message comprises the basic rescue information, the number of passengers and the accident type label.
  9. 9. A vehicle control system, characterized by comprising: the acquisition module is used for acquiring driving state information of one or more dimensions of a driver; the analysis module is used for analyzing the driving state information by using a machine learning model to obtain a driving incapacitation risk index of the driver, wherein the driving incapacitation risk index is used for representing the probability or degree of incapacitation of the driver for safe driving; and the control module is used for taking over the vehicle driven by the driver under the condition that the driving disability risk index is larger than a preset threshold value.
  10. 10. A vehicle comprising a computer program, characterized in that a controller of the vehicle is adapted to execute the vehicle control method according to any one of claims 1 to 8.
  11. 11. A nonvolatile storage medium, characterized in that the nonvolatile storage medium includes a stored program, wherein the program, when executed, controls a device in which the nonvolatile storage medium is located to execute the vehicle control method according to any one of claims 1 to 8.
  12. 12. An electronic device comprising a memory and a processor for executing a program stored in the memory, wherein the program is executed to perform the vehicle control method according to any one of claims 1 to 8.
  13. 13. A computer program product comprising a computer program, characterized in that the computer program, when executed by a processor, implements the vehicle control method of any one of claims 1 to 8.

Description

Vehicle control method, system and electronic equipment Technical Field The application relates to the field of vehicle safety early warning and takeover, in particular to a vehicle control method, a system and electronic equipment. Background In the related intelligent driving assistance technology, the response of a vehicle to the abnormal state of a driver is generally remained on the early warning and prompting level, and when the driver loses the control capability due to sudden cardiovascular and cerebrovascular diseases, serious fatigue, consciousness loss or sudden physiological stress reaction, the vehicle can only send out a take over request through the modes of audible and visual alarm, seat vibration or instrument panel prompt and the like, so that the accurate judgment and active intervention mechanism for judging whether the driver really loses the control capability are not lacking. The related scheme is highly dependent on the capability of timely response of a driver after receiving a prompt, once the driver cannot perform taking over operation due to fuzzy consciousness, weak limbs or complete unconsciousness, the vehicle keeps the original running state, continues to advance at a high speed along the original track, cannot automatically decelerate, change lanes or stop, and is extremely easy to cause secondary accidents such as rear-end collision, side turning or road rushing out, and serious personal injury and public safety risks are caused. In view of the above problems, no effective solution has been proposed at present. Disclosure of Invention The application provides a vehicle control method, a vehicle control system and electronic equipment, which at least solve the technical problem that a vehicle is out of control continuously in a non-response state because the related technology cannot actively take over the control of the vehicle when a driver is suddenly disabled. According to one aspect of the application, a vehicle control method is provided, and the vehicle control method comprises the steps of obtaining driving state information of one or more dimensions of a driver, analyzing the driving state information by using a machine learning model to obtain a driving incapacity risk index of the driver, wherein the driving incapacity risk index is used for representing the probability or degree of incapacity of the driver to safely drive, and taking over a vehicle driven by the driver under the condition that the driving incapacity risk index is larger than a preset threshold value. Optionally, the driving state information of the multiple dimensions comprises physiological signals, visual behavior signals, audio signals and driving operation trigger signals; the machine learning model is obtained through training by acquiring sample data, wherein the sample data carries labels marked with incapacitation states, the sample data comprises synchronously acquired contact type physiological and electric characteristic data, non-contact type visual behavior characteristic data, audio characteristic data and driving control discrete signals, carrying out time domain alignment on the contact type physiological and electric characteristic data, the non-contact type visual behavior characteristic data, the audio characteristic data and the driving control discrete signals based on a unified time reference, carrying out synchronous slicing on the aligned data according to preset time length to obtain multi-mode synchronous original data fragments, extracting physiological characteristic vectors, visual characteristic vectors and audio characteristic vectors from the multi-mode synchronous original data fragments, carrying out space mapping on the physiological characteristic vectors, the visual characteristic vectors and the audio characteristic vectors to obtain multi-mode mapping characteristics, constructing a driving state space-time correlation tensor according to control deviation values of driving control discrete signals corresponding to the multi-mode synchronous original data fragments, determining the characteristic strength of corresponding dimensions of the modal characteristic components according to the change rate of the driving state space-time correlation tensor, calculating the characteristic strength corresponding to the modal space-time correlation tensor, taking the weighting factors as a fusion characteristic gradient loss prediction coefficient as a fusion weighting function after the most weighting factor of the preset contribution weight is reduced, until the parameter convergence of the machine learning model is completed. The method comprises the steps of obtaining acceleration characteristics representing vibration intensity of a vehicle and visual quality characteristics representing illumination environment change, obtaining driving behavior information and external environment information, carrying out motion artifact correlation analysis on physiological c