Road Object Detection Based On Deep Learning

With the vigorous development of the automobile industry in recent years,cars have become the main means of transportation in modern society and bring a lot of convenience to people's daily travel.However,the increasing number of car ownership has made traffic safety problems and urban congestion problems more and more serious.As a branch of artificial intelligence,deep learning has developed rapidly in the era of Internet and big data.Road object detection is an important part of assisted driving technology system.This paper uses the convolutional neural network of deep learning to realize road object detection.The main contents of this paper are as follows:(1)Capture images from driving videos and annotate these images with the Labellmg annotation tool to obtain the location and category information of the objects in these images.Then,a road object detection dataset is created.By analyzing the advantages and disadvantages of the two-stage object detection algorithm and the one-stage object detection algorithm based on the deep convolutional neural network and weighing the accuracy and speed requirements of the road object detection task,the SSD network of multi-scale feature map prediction combined with non-maximum suppression algorithm and hard negative sample mining is introduced into the road object detection task.The mAP of the network is 76.6% and the time to detect one picture is 0.35 s.(2)In order to speed up the detection speed of the SSD network and realize the real-time detection requirements for road object,the MobileNet-SSD road object detection network is built by using MobileNet instead of VGG feature extraction network in SSD network.MobileNet is a lightweight convolutional neural network.This method reduces the number of parameters and the amount of calculations in the SSD network and doubles the detection speed of the network.(3)In order to improve the detection accuracy of the MobileNet-SSD network,the improved MobileNet network is used as the feature extraction network in SSD model.After that,the detection accuracy of car object is improved by 2.1%,the detection accuracy of cyclist object is improved by 4.8%,the detection accuracy of person object is improved by 7.3% and the mAP is improved by 4.7%.Then,the K-means algorithm is used to cluster the width and height of the object in the training dataset and the aspect ratio of the anchor frame is corrected.So that the anchor frame shape of the network can better match the road object.After correcting the anchor frame of the network,the detection accuracy of cyclist object is improved by 2.1%,the detection accuracy of person object is improved by 1.5% and the mAP is improved by 1.1%.(4)The road monitoring video was tested with the road object detection network based on deep leaning designed in this paper.The network runs stably at twice the speed of the SSD network and the accuracy of the network also meets the requirement.The test results show that the detection network designed in this paper can meet both the functional and performance requirements in practical applications and has certain practical value.