Research On Algorithm Of Vehicle Blind Zone Hazard Detection Based On Lightweight Network

Auto A pillar refers to the pillar connecting the door and roof on both sides of the front windshield of the car.It is an important factor to ensure the strength of the car body and directly affects the safety of the driver when the car is hit.Therefore,the design of auto A pillar has very strict specifications.The presence of A pillar will block the driver’s vision,thus affecting the driver’s judgment and causing serious safety hazards.In view of the current situation of traffic accidents caused by the blind area formed by the A pillar,it is more and more important to apply artificial intelligence technology to the A pillar.Firstly,this paper proposes to design a lightweight network based on YOLOv4,which is used for the detection of vehicle blind zone hazards;Due to the special space of the blind area display screen and the camera installation position,in this paper,the three-dimensional coordinate system is used to calculate the three-dimensional coordinates of the point space.Then,an algorithm is proposed to calculate the blind area using the center point of the human eye.The specific research contents are as follows:(1)In this paper,the target detection algorithm YOLOv4 is optimized on the network structure.The model is compressed by replacing the backone of the original model structure with Mobile net.The original ordinary convolution of the model is replaced by the deep separable convolution with less parameters to further compress the model.Replace the detection head part of the model with the Decoupled head,and remove the two 3 in the original Decoupled head × 3 Convolution,GIo U is used for regression head branch.Through experimental comparison,the optimized model m AP in this paper is 87.0%,1.4 points higher than the original model YOLOv4,the weight size is only 1/4 of the original model,and the detection rate reaches 56 FPS.(2)This paper analyzes the blind area in the driver’s field of vision when the vehicle is running normally.It is assumed that the blind corner area where the A-pillar is located will not change during driving.That is,the assumption that the position of driver’s head and A-pillar is relatively fixed is put forward.Based on this assumption,the camera is used to take the head picture of the current position.The light weight network designed above is used to identify and locate the human eye,calculate the coordinates of the center point of the human eye,and calculate the blind area by simulating the vision.(3)This paper sets up the feasibility of hazard detection of the external blind area through the analysis and design of the lightweight network.When the blind area scene is detected to be dangerous,wake up the A-pillar blind area elimination system,and display the blind area in real time,which well solves the driver’s visual fatigue,distraction and other problems caused by the constant brightness of the A-pillar display.