Road Object Detection And Road Condition Recognition Technology For Intelligent Assisted Driving

In modern society,as one of the most important transportation tools in people’s daily life,automobile provides us with great convenience.But it brings a series of hidden problems at the same time,such as driving safety,riding comfort and so on.In particular,the number of road traffic accidents has been increasing year by year.Intelligent assisted driving is a new way to address the problems above and has become a hot spot in today’s technology.This paper mainly focuses on the task of intelligent assisted driving,and studies road object detection and road condition recognition,aiming at improving the ability of automobile assisted driving system to find targets and probe road conditions.Therefore,it has very important theoretical and application value.The main work of this paper is as follows:Firstly,the development process and research status of both object detection and road condition recognition are introduced in detail,and the related classical algorithms are analyzed and compared in this paper.Then,an improved algorithm model of YOLO v3 for road object detection is proposed.Considering the problem of limited memory space and computing power of automobile embedded devices,this paper improves the YOLO v3 by network slimming and loss function replacement.Thirdly,an improved algorithm model of Darknet-53 for road condition recognition is proposed.The road condition recognition,which takes the vehicle vibration amplitude level as labels and is different from the traditional counterpart particularly,is regarded as an image classification problem for the first time.And the algorithm is improved from two directions,namely reducing the number of output channels in convolution layers and adding a full connection layer.For road condition recognition dataset,a dataset collection scheme with the automobile vibration amplitude levels as labels is proposed,which is used to collect road condition image and its amplitude to the automobile.And the collection,calibration and augment of the dataset is completed as well.Finally,for the object detection model based on YOLO v3 algorithm and the road condition recognition model based on Darknet-53 network both proposed in this paper,relevant experiments are carried out on the augmented BDD100 K public dataset and the road condition recognition dataset collected in this paper respectively.The experiment results show that compared with the corresponding original model,the proposed road object detection model reduces the size by about 77.8%,increases the running speed by about 53.4 fps and increases the m AP-50 by about 2.6%,and the proposed road condition recognition model reduces the size by about 74.3%,increases the running speed by about 1.7 fps and increases the classification accuracy by about 5.98%.Therefore,the proposed models meet the requirements of the scenario in this paper because of the good comprehensive performance,and have certain reference and application value.