Preceding Object Detection For Vehicle Based On Deep Learning

With the development of the times and the advancement of technology,traveling by motor vehicle has become a way of people's daily life,and the accompanying intelli-gent transportation system has become the object of development for governments and major technology companies in various countries.Due to the breakthrough development of computer hardware technology,various algorithms based on deep neural networks are no longer limited by computing resources.As an important part of intelligent trans-portation systems,the problem of object detection in assisted driving systems has be-come a hot research field.The object detection problem in the assisted driving system is mainly divided into two sub-tasks:image classification and obeject location detection.The general applica-tion scenario is the road image in front of the vehicle.Due to the unique advantages of the convolutional neural network's own structure when processing image data,the in-dustry often uses various types of convolutional neural networks as road target detection methods.This paper uses these algorithms from the perspective of feature extraction,feature classification and position regression.The classification is introduced,and the advantages of using YOLO series algorithms are explained.In order to solve the problem of object detection in the road ahead video,this paper uses the YOLO-tiny algorithm to detect the target in front of the road,but considering the problem of high real-time but insufficient detection accuracy when the YOLO-tiny algorithm is performing the object detection task,in order to meet For the real-time and accuracy requirements of the detection algorithm in the assisted driving system,this pa-per has redesigned the network model of YOLO-tiny.While deepening the number of network layers,a 1*1 convolution layer is intro-duced to simplify the parameters of the network,so that the redesigned network can ex-tract deeper features,and at the same time make the real-time reduction of the algorithm still within the acceptable range;in order to further improve the detection ability of the network,this paper tried different numbers of The prior frame is re-clustered,and the effect of the size of the prior frame from each cluster is analyzed and compared.In order to solve the problem of missed detection of continuous moving objects when the object detection algorithm detects video targets,a object tracking module based on the Kalman filter algorithm is added to the object detection process,and each detected object is continued for a period of time.For tracking,the obtained detection edge frame and the tracking edge frame are optimally matched using the Hungarian al-gorithm to improve the target loss phenomenon during YOLO-tiny detection,and make full use of the correlation between the upper and lower frame information in the road video.Finally,several comparative experiments are designed in this paper.Through the analysis of the performance indicators and the experimental results in consecutive im-age frames of the same video stream,the object detection algorithm described in this pa-per is verified to have good detection accuracy and real-time performance.