Research On Pedestrian Target Detection And Ranging Technology Based On Deep Learning

In recent years,with the increase of car ownership,people not only enjoy the convenience of life brought by cars,but also suffer from traffic accidents.In order to suppress the frequent traffic accidents and ensure the safety of life and property,timely perceive the traffic environment,identify the pedestrians in front and measure the relative distance between the target and our vehicle,so as to provide help and support for the driver,it has become a research hotspot of the current advanced assistant driving system.The traditional pedestrian detection algorithm can not avoid the problems caused by lighting,weather,road conditions and pedestrian itself,which leads to frequent problems in actual operation.How to deal with it better is the current dilemma to be solved.This paper tries to take the pedestrians in the traffic environment as the research subject,and explores the pedestrian target detection and ranging technology based on deep learning.Through a series of methods such as improving the target detection algorithm based on deep learning and optimizing the ranging algorithm,it aims to construct a relatively perfect pedestrian target ranging system.Relying on the current hot deep learning technology,aiming at the difficulty that the current detection algorithm is difficult to ensure the real-time operation of the system at the same time of high-quality detection,this paper proposes a high detection accuracy and faster Yolov4-P target detection algorithm based on the general target detection algorithm Yolo v4,which uses the techniques of equal scale image scaling and focus slice reconstruction to improve the accuracy of the algorithm for small target pedestrians Finally,on the verification set of BDD100K data set,the AP₅₀value of Yolov4-P algorithm reaches 51.86%,the average time of detecting a picture is only 16 ms,and the speed is nearly 25%higher than Yolo v4.Due to the diversity of human posture in driving environment,it is impossible to use data regression modeling method to build a mathematical model for pedestrian distance measurement.In this paper,a geometric model of roll angle is proposed based on the roll angle correction.Combined with the target detection algorithm,the pedestrian target ranging system based on deep learning is realized.The running speed of the pedestrian target ranging system designed in this paper reaches 62.5 frames per second,and the average ranging error is less than 6%in the longitudinal distance of 60 meters,and the pitch angle can be corrected by calculating the vanishing point of the road,which can meet the real-time pedestrian target detection and ranging requirements of the advanced auxiliary driving system.