Research On Vehicle Detection And Driving Intention Recognition Based On Machine Vision

Automatic driving is currently the most popular research area in the automobile industry.To be safe,the self-driving car must be able to sense the surrounding environment and avoid collisions with obstacles,especially the front vehicles.At present,the research on environmental perception is mainly based on the information obtained by sensors such as camera,millimeter wave radar and laser radar.Vehicles ahead can be detected using the environment-aware technology.However,if automatic driving technology really wants to replace a human driver,it is necessary to ensure ride comfort and fuel economy on the basis of safety.This requires autonomous vehicles to understand road scenes like humans,thus providing a basis for driving decisions.Based on this,this paper studies the front vehicle detection and driving intention recognition method based on machine vision.First of all,this paper improves the model of the latest deep learning algorithm YOLO V3 in the field of target detection to detect the front vehicles.The improved model has improved detection speed and better real-time performance when the accuracy is constant.Secondly,according to the imaging principle of the camera,a method of measuring the distance and the vehicle width based on monocular ranging is proposed.We calibrate the camera to get the camera’s internal and external parameters,combined with the pixel information of front vehicles obtained by the front vehicle detection and the height of the camera from the ground,the distance and width of the front vehicle can be calculated.Then,the detected tail region of the front vehicle is extracted under the HSV color space,which is classified by the deep learning network named VGG,thereby realizing the recognition of the front taillights and learning the driving intention.Finally,the method proposed in this paper is verified by experiments.The experimental results show that using this method,the accuracy of the front vehicle detection is 88.05%.In the range of 30 meters,the accuracy of the distance measurement is not less than 94.69% and the vehicle width measurement accuracy is not less than 97.22%.Tail light recognition accuracy rate reached 86.7%.