Search

CN-121989933-A - Transport vehicle anti-collision early warning and braking control system based on AI vision fusion

CN121989933ACN 121989933 ACN121989933 ACN 121989933ACN-121989933-A

Abstract

The invention provides an AI vision fusion-based transportation vehicle anti-collision early warning and braking control system, which is characterized in that a multi-mode sensing device is deployed to acquire full-dimensional AI vision data of a transportation vehicle to obtain multi-mode data, images, speeds, heat radiation and vehicle load data can be accurately acquired, the full-dimensional capturing environment and vehicle states are firmly built for subsequent processing, multi-source fusion data are obtained by asynchronously fusing the multi-mode data, data integration can be rapidly completed, synchronous fusion delay is reduced, meanwhile, privacy leakage risks are avoided when data uploading, a dynamic risk field is established by establishing the multi-source fusion data, risk assessment is carried out on the vehicle states in the dynamic risk field, grading early warning and braking decisions are obtained according to risk assessment results, the dynamic risk assessment is combined with road conditions and load adjustment grades, the grading early warning and the braking decisions are matched with different scenes, and safety and operation efficiency are balanced.

Inventors

  • WANG DI
  • SUN LIANQING

Assignees

  • 山西云启帮科技有限公司

Dates

Publication Date
20260508
Application Date
20260205

Claims (10)

  1. 1. Transport vehicle anticollision early warning and braking control system based on AI vision fuses, characterized by comprising: the sensing layer is used for acquiring full-dimension AI visual data of the transport vehicle based on the deployed multi-mode sensing device to obtain multi-mode data; The edge computing layer is used for asynchronously fusing the multi-mode data to obtain multi-source fusion data; The decision control layer is used for establishing a dynamic risk field based on the multi-source fusion data, carrying out risk assessment on the state of the vehicle in the dynamic risk field, and obtaining grading early warning and braking decisions according to the risk assessment result; And the execution layer is used for controlling the braking system to perform cooperative work according to the braking decision and performing collision prediction in real time, and providing steering auxiliary parameters when the fact that collision cannot be avoided is determined.
  2. 2. The AI vision fusion-based anti-collision early warning and braking control system for a transport vehicle according to claim 1, further comprising a self-learning optimization module for summarizing transport vehicle driving data, and periodically performing iterative updating on the asynchronous fusion, the dynamic risk assessment mechanism and the auxiliary parameters based on the driving data.
  3. 3. The AI visual fusion-based anti-collision warning and braking control system for a transport vehicle of claim 1, wherein the perception layer comprises: the visual camera is used for collecting a two-dimensional image of a scene in front of the transport vehicle and outputting the outline, texture characteristics and relative distance data of the obstacle; millimeter radar wave, which is used to collect the speed, azimuth angle and motion trail data of obstacle; the infrared imager is used for starting in a low-light scene and capturing the heat radiation characteristics of the obstacle; the environment sensor is used for collecting road adhesion coefficient, illumination intensity, rainfall and real-time state data of the vehicle.
  4. 4. The AI vision fusion-based anti-collision warning and braking control system for a transport vehicle of claim 1, wherein the edge calculation layer comprises: the rapid processing unit is used for carrying out light neural network processing on the single-path visual image in the multi-mode data to obtain two-dimensional processing data; the accurate processing unit is used for carrying out time space alignment on the binocular vision image, the infrared thermal imaging image and the laser radar point cloud to obtain three-dimensional processing data; and the asynchronous fusion unit is used for correlating the two-dimensional processing data with the three-dimensional processing data, adopting the two-dimensional processing data to extrapolate when the three-dimensional processing data is delayed, and updating the extrapolation result immediately after the three-dimensional processing data is acquired.
  5. 5. The AI visual fusion-based anti-collision early warning and braking control system for a transport vehicle of claim 1, wherein the decision control layer comprises: The risk field establishing unit is used for establishing a static obstacle risk field, a dynamic target prediction risk field and a road structure risk field based on the multi-source fusion data, and superposing the static obstacle risk field, the dynamic target prediction risk field and the road structure risk field to obtain a dynamic risk field; the track analysis unit is used for predicting a plurality of short-term tracks of the transport vehicle based on the current speed of the transport vehicle, the steering wheel angle and the intention of a driver, acquiring a longitudinal predicted position of each short-term track along the direction of a central line of a road at any time and a transverse predicted position of each short-term track perpendicular to the direction of the central line of the road based on a dynamic risk field, determining a risk concentration value, multiplying the risk concentration value of the position of each instant by the instant speed of the transport vehicle in a future time window of each short-term track, continuously accumulating all instant results to obtain accumulated risk exposure, and extracting the maximum concentration value of the risk concentration value of each short-term track and the occurrence time point; The vector establishing unit is used for acquiring the accumulated risk exposure and the change rate of the maximum concentration value along with time, integrating the state of a driver, the state of a vehicle, a plurality of short-term tracks, the accumulated risk exposure, the maximum concentration value, the occurrence time point and the change rate, and obtaining a risk state vector; the grading early warning unit is used for comparing the risk state vector with an early warning rule threshold value and determining grading early warning according to a comparison result; And the braking decision unit is used for taking a gradient braking control mechanism matched with the dynamic risk field as a braking decision when the grading early warning shows that the driver is not involved in the reaction.
  6. 6. The AI-vision-fusion-based anti-collision early warning and braking control system for a transport vehicle of claim 5, wherein the risk field establishment unit comprises: the static risk field building unit is used for giving basic risk weight based on physical properties of the obstacle, giving dynamic risk weight based on specific characteristics of the obstacle changing along with the change of environmental parameters, generating a single obstacle risk influence area based on the basic risk weight, the dynamic risk weight and the obstacle characteristics, and superposing all the single obstacle risk influence areas to obtain a static obstacle risk field; The target risk field establishing unit is used for predicting a plurality of future motion tracks through a deep learning model based on the track of the dynamic target, determining the probability of each future motion track, establishing a probability cloud of the dynamic target along with time, generating a virtual repulsive force vector field proportional to the predicted kinetic energy of the dynamic target, convoluting the virtual repulsive force vector field with the probability cloud, and generating a dynamic target predicted risk field by combining the type gain of the dynamic target; The road risk field establishment unit is used for acquiring road curvature and a real-time estimated vehicle lateral adhesion coefficient, determining lateral instability risk, generating an additional risk gradient proportional to the square of curvature and speed at the outer side of a curve, setting a quadratic attraction potential well at the center line of a lane based on the lateral instability risk and the additional risk gradient, and setting an exponential rejection potential barrier at the lane line and the road boundary to obtain a road structure risk field; the risk field integration unit is used for defining longitudinal distance, transverse offset and time by taking a front road center line as a reference, establishing a risk field coordinate system, and configuring a static obstacle risk field, a dynamic target prediction risk field and a road structure risk field into the risk field coordinate system to obtain a dynamic risk field.
  7. 7. The AI vision fusion-based anti-collision early warning and braking control system for a transport vehicle of claim 6, further comprising a risk field calibration unit that calibrates the dynamic risk field in real time based on a three-level adaptive calibration model; The first layer in the three-level self-adaptive calibration model is based on vehicle physical state calibration: Acquiring a vehicle load, and calibrating a dynamic risk field in real time according to a rule that a risk area of a static obstacle is enlarged and a remote influence range is relatively small when the vehicle load is heavy; acquiring a road surface adhesion coefficient, and calibrating a dynamic risk field in real time according to the same rule that the curve is more dangerous under low adhesion; The second layer in the three-layer self-adaptive calibration model is based on the calibration of the driver behavior portraits; based on a risk preference coefficient, a reaction time model and a trust degree attenuation factor of a driver, calibrating a dynamic risk field in real time; the third layer in the three-layer self-adaptive calibration model is based on scenerization dynamic decision calibration; And establishing a dynamic early warning threshold curve and a dynamic braking threshold curve to calibrate the dynamic risk field in real time.
  8. 8. The AI vision fusion-based anti-collision warning and braking control system for a transport vehicle of claim 5, wherein the hierarchical warning unit comprises: The first-level early warning unit is used for judging that the first-level early warning is performed when the accumulated risk exposure exceeds a dynamic baseline threshold value determined based on the driver files and road conditions and the maximum concentration value is not in an emergency area; the secondary early warning unit is used for judging secondary early warning when the maximum concentration value is larger than the primary warning threshold value and the accumulated risk exposure shows an ascending trend; And the three-level early warning unit is used for judging three-level early warning when the maximum concentration value is greater than the high-level warning threshold value.
  9. 9. The transport vehicle anti-collision early warning and braking control system based on AI visual fusion according to claim 5, wherein the gradient braking control mechanism in the braking decision unit is specifically as follows: a pre-braking stage, namely, once braking arbitration is carried out, pre-filling a braking system immediately; The proportional following stage is that the target braking force is in direct proportion to the gradient of the current moment and is limited by the current road adhesion coefficient and the vehicle load; And in the full braking stage, if the risk is suddenly increased, the vehicle enters a full braking mode, and the maximum safety braking force is applied.
  10. 10. The AI visual fusion-based anti-collision warning and braking control system for a transport vehicle of claim 1, wherein the execution layer comprises: the collision prediction unit is used for judging whether collision can be avoided by means of braking according to the real-time vehicle dynamics state and the dynamic risk field; And the parameter determining unit is used for generating steering auxiliary parameters according to the lateral space data acquired in real time, wherein collision cannot be avoided by virtue of braking.

Description

Transport vehicle anti-collision early warning and braking control system based on AI vision fusion Technical Field The invention relates to the technical field of vehicle control, in particular to an anti-collision early warning and braking control system for a transport vehicle based on AI vision fusion. Background With the large-scale development of logistics industry, the running safety of a transport vehicle directly relates to personnel life, goods property and road passing efficiency, the existing transport vehicle anti-collision system is based on a single vision sensor or a basic fusion scheme of laser radar and vision, and has the limitations that firstly, the system is poor in adaptability to complex environments, obstacles are easy to be mistakenly identified and missed to identify in severe weather such as strong light, shadow, rain and fog, particularly, the distinguishing capability of virtual obstacles (such as road surface reflection and identification shadow) is insufficient, so that mistaken early warning or braking is caused, secondly, the dynamic target track prediction is single, collision risk is judged only based on distance and speed parameters, braking delay and goods inertia influence caused by the heavy load characteristic of the physical distribution vehicle are not combined, risk assessment accuracy is low, thirdly, a braking control strategy is solidified, a cutter-cut mode of early warning emergency braking is adopted, different road conditions (high speed, national road, factory areas) and goods state are not adapted, and other secondary risks such as goods dumping and vehicle rollover are easy to be caused, fourthly, the system is poor in self-adaptive learning capability, model parameters cannot be optimized according to historical data of different transport scenes, and the system is poor. In order to solve the above problems, a need exists for an anti-collision early warning and braking control system that integrates multi-modal visual information, adapts to the characteristics of logistics vehicles, has dynamic decision making and self-optimizing capabilities, breaks through the conventional framework of the prior art, and improves the safety protection reliability in complex scenes. Disclosure of Invention The invention provides an AI vision fusion-based anti-collision early warning and braking control system for a transport vehicle, which is used for solving the problems in the background technology. An AI vision fusion-based transport vehicle anti-collision early warning and braking control system, comprising: the sensing layer is used for acquiring full-dimension AI visual data of the transport vehicle based on the deployed multi-mode sensing device to obtain multi-mode data; The edge computing layer is used for asynchronously fusing the multi-mode data to obtain multi-source fusion data; The decision control layer is used for establishing a dynamic risk field based on the multi-source fusion data, carrying out risk assessment on the state of the vehicle in the dynamic risk field, and obtaining grading early warning and braking decisions according to the risk assessment result; And the execution layer is used for controlling the braking system to perform cooperative work according to the braking decision and performing collision prediction in real time, and providing steering auxiliary parameters when the fact that collision cannot be avoided is determined. Preferably, the self-learning optimization module is further used for summarizing the running data of the transport vehicle, and periodically carrying out iterative updating on the asynchronous fusion and the dynamic risk assessment mechanism and the auxiliary parameters based on the running data. Preferably, the sensing layer includes: the visual camera is used for collecting a two-dimensional image of a scene in front of the transport vehicle and outputting the outline, texture characteristics and relative distance data of the obstacle; millimeter radar wave, which is used to collect the speed, azimuth angle and motion trail data of obstacle; the infrared imager is used for starting in a low-light scene and capturing the heat radiation characteristics of the obstacle; the environment sensor is used for collecting road adhesion coefficient, illumination intensity, rainfall and real-time state data of the vehicle. Preferably, the edge calculation layer includes: the rapid processing unit is used for carrying out light neural network processing on the single-path visual image in the multi-mode data to obtain two-dimensional processing data; the accurate processing unit is used for carrying out time space alignment on the binocular vision image, the infrared thermal imaging image and the laser radar point cloud to obtain three-dimensional processing data; and the asynchronous fusion unit is used for correlating the two-dimensional processing data with the three-dimensional processing data, adopting the two-dimensio