Search

CN-121989926-A - Active anti-collision method and device for vehicle

CN121989926ACN 121989926 ACN121989926 ACN 121989926ACN-121989926-A

Abstract

The invention discloses a method and a device for actively preventing a vehicle from collision with a person and the vehicle, and belongs to the technical field of active safety of vehicles. The method comprises the steps of comparing the real-time position of a vehicle to be driven with a preset electronic fence database, when the vehicle is judged to enter a critical area with dense people flow, forcedly activating a protection system, after the system is activated, running a special peripheral personnel recognition model to strengthen perception and accurately recognize targets such as peripheral pedestrians, children, baby carriages and wheelchairs, synchronously fusing target information with the state of the vehicle, dynamically evaluating collision risk level, and finally executing graded active control from acousto-optic warning, partial braking, full-force automatic braking to cooperative steering emergency avoidance according to the risk level and combining with operation feedback of a driver. Aiming at the defects that the existing active safety system is insufficient in protection of personnel outside the vehicle and can be manually closed, the intelligent active protection which is unshielded in a key area and takes pedestrians as a core is realized, the risk of malignant accidents caused by malicious collision or out-of-control of the vehicle is effectively reduced, and the system is easy to deploy based on the existing vehicle platform.

Inventors

  • FENG GUANGLANG

Assignees

  • 北京辉羲智能科技有限公司

Dates

Publication Date
20260508
Application Date
20260226

Claims (10)

  1. 1. A method for actively preventing a collision of a vehicle, comprising the steps of: S1, judging a key area, namely acquiring positioning information of a vehicle in real time, and comparing the positioning information with a preset or real-time updated electronic fence database; S2, a special sensing step of calling and running a preset surrounding personnel identification model in response to the forced activation signal, wherein the surrounding personnel identification model receives data from a vehicle-mounted environment sensor and is specially used for identifying and outputting type, position and motion state information of pedestrians and related targets around the vehicle; s3, risk analysis step, namely synchronously acquiring real-time running state parameters of the vehicle, and calculating potential collision risk level in the current running direction of the vehicle by combining the target information output by the special perception step; and S4, performing graded active intervention operation according to the risk grade output by the risk analysis step, wherein the graded control operation comprises at least one of giving early warning, implementing partial braking, implementing full-force automatic braking, and controlling the steering of the vehicle to avoid collision.
  2. 2. The method for actively preventing collision of a vehicle according to claim 1, wherein the surrounding personnel identification model is constructed based on a deep learning architecture combining BEV and a transducer, and the processing flow comprises the steps of inputting multi-path camera data into a shared backbone network to extract characteristics, converting the characteristics into a unified BEV space through an attention mechanism, fusing multi-sensor data and performing time sequence modeling, and finally outputting dynamic sensing results of at least one target of pedestrians, children, riders, strollers and wheelchairs.
  3. 3. The vehicle active collision avoidance method according to claim 1, characterized in that in the step of hierarchical control, specifically comprising: calculating a safe distance threshold D based on the current vehicle speed and the target type; calculating the relative distance d between the vehicle and the target in real time; And dividing the risk state into multiple stages according to the comparison relation between the relative distance D and the safety distance threshold D and the operation feedback of the driver after early warning, and correspondingly executing control instructions of different stages.
  4. 4. A method of actively preventing a collision of a vehicle according to claim 3, in which the risk status comprises at least: A first-level dangerous state (D0) for controlling the vehicle to emit an audible and visual warning when the relative distance D approaches the safe distance threshold D; a secondary dangerous state (D1) for controlling the vehicle to perform partial automatic braking when the vehicle is in the primary dangerous state (D0) and the driver does not respond effectively; A third dangerous state (D2) for controlling the vehicle to execute automatic emergency braking of the maximum deceleration when the risk is further raised; And a four-stage dangerous state (D3) for planning and controlling the vehicle to turn to avoid the obstacle while braking when judging that the braking cannot avoid collision.
  5. 5. The method for actively preventing collision of a vehicle according to claim 1, in which the surrounding personnel recognition model operates independently of an original automatic driving or auxiliary driving perception model of the vehicle, and functions thereof do not include lane line recognition and global path planning.
  6. 6. The method of actively preventing a collision of a vehicle according to claim 1, in which the electronic fence database receives real-time updates of a traffic management platform through the internet of vehicles or is provided by vehicle-mounted navigation map data.
  7. 7. A vehicle active collision avoidance device for implementing the method of any one of claims 1 to 6, the device being integrated in the vehicle's electronic-electrical architecture, comprising: the positioning and area judging module is used for acquiring vehicle positioning information, comparing the vehicle positioning information with the electronic fence database and outputting a forced activation signal when the vehicle enters a key area; The special perception module is connected with the positioning and area judging module and is used for operating the surrounding personnel identification model after receiving the forced activation signal and processing the environment sensor data to identify the surrounding personnel targets; the comprehensive analysis module is connected with the special perception module and the vehicle bus, and is used for fusing target information and real-time state parameters of the vehicle and evaluating collision risk levels; the grading decision and control module is connected with the comprehensive analysis module, and is used for generating a control instruction according to the risk grade and preset grading logic and issuing the control instruction to the executors of the vehicle body domain, the chassis domain and the power domain.
  8. 8. The vehicle active collision avoidance device of claim 7 wherein the special awareness module, the comprehensive analysis module, and the hierarchical decision and control module are deployed in a self-driving domain controller, a cabin domain controller, or a combination of both of the vehicle.
  9. 9. The active collision avoidance device of claim 7 wherein the control commands generated by the hierarchical decision and control module are used to control at least one of a dual flash, horn, brake by wire system, electric power steering system, and power system of the vehicle.
  10. 10. A vehicle, characterized in that a vehicle active collision avoidance device as claimed in any one of claims 7 to 9 is mounted.

Description

Active anti-collision method and device for vehicle Technical Field The invention belongs to the technical field of active safety and intelligent driving of vehicles, and particularly relates to an active anti-collision method and an active anti-collision device for vehicles, which are particularly suitable for identifying surrounding pedestrian states in a people stream dense area, predicting collision risks and implementing active intervention so as to reduce risks that vehicles are maliciously used for collision attack or malignant accidents are caused by misoperation of drivers and system faults of the vehicles. Background As urban population density increases and vehicle retention continues to rise, public transportation safety presents a serious challenge. . The traditional vehicle has the defect of safety control that once the vehicle is maliciously operated, the vehicle lacks a forced safety intervention mechanism for the driving behavior of the vehicle, and the vehicle is very easy to be used as a damaging tool. On the other hand, it is common that a vehicle is out of control accident due to improper operation of a driver, sudden physical abnormality, malfunction of hardware and software of the vehicle, and the like. Accidents such as out-of-control crashes of the south-Ning SUV in 2024 and suspected brake failure in Tesla, chaozhou all show that tragedy occurrence is difficult to completely avoid by just relying on the immediate response of a driver in a complex traffic environment. The conventional countermeasure mainly comprises two types, namely passive defense measures, such as physical roadblocks, reinforced police inspection and the like, which are arranged in key places, a large amount of manpower and material resources are needed to be input, the normal traffic flow is disturbed, the protection range and effect are limited, and active safety technologies of vehicles, such as an anti-lock brake system (ABS), an Electronic Power Steering (EPS), automatic Emergency Braking (AEB) and the like. The AEB system detects a front obstacle through a sensor such as a radar or a camera, and automatically applies braking when judging that there is a collision risk, which has become an important technology for improving driving safety. However, the existing AEB system still has the following limitations that firstly, the system is designed to protect passengers in a vehicle, the perception and protection capability of traffic participants such as outsides of vehicles and non-motor vehicles are insufficient, secondly, the effective working speed range of the system is limited, the recognition capability of low-speed, static or special-form targets (such as children, baby carriages and wheelchairs) is weak, the recognition distance is short, the omission ratio is high, thirdly, most AEB systems cannot be activated under partial working conditions, for example, the functions of the passengers are limited when the passengers are not fastened with safety belts, fourthly, the system is usually a selectable option of a driver, and forced starting cannot be guaranteed under a key scene. In recent years, with the development of artificial intelligence and automatic driving technology, a perception scheme based on fusion of vision and multiple sensors has been advanced in understanding the environment of a vehicle. The main stream automatic driving system mostly adopts an end-to-end or Visual Language Model (VLM), and the main points of the VLM are lane line identification, path planning and common motor vehicle detection, so that the accurate identification of targets of pedestrians, particularly special groups of people, is still insufficient, the system can be manually closed by a driver, and the forced intervention capability is lacked in safety critical scenes. Therefore, a system capable of enhancing surrounding pedestrian perception, particularly forcible starting in a dense pedestrian flow area, and performing real-time risk assessment and active control according to pedestrian status and vehicle dynamics is needed to make up for the defect of the prior safety technology in terms of protecting outside personnel and improve the driving safety of public space. Disclosure of Invention The invention aims to overcome the defects in the prior art that (1) the existing Automatic Emergency Braking (AEB) and other systems take the protection of personnel in a vehicle as priority, have limited perception capability on pedestrians outside the vehicle, especially have low recognition rate on small targets and special targets, (2) the existing systems have limited working conditions (such as a vehicle speed range and a safety belt state) and can be manually closed by a driver, and cannot ensure forced intervention in a critical safety scene, (3) a main stream automatic driving perception model takes lane keeping and path planning as cores, and has no design key point on the special recognition and tracking capability o