Research And Development Of Forward-looking Algorithm In ADAS System

Driving safety is an eternal theme for the development of automobile traffic.With the rapid increase of car ownership,the incidence of traffic accidents remains high,and traffic safety issues have received widespread attention.Unintentional lane departure and driver fatigue driving are important causes of accidents,and lane departure warning systems and front collision warning systems in advanced driving assistant system can effectively avoid accidents caused by these reasons.Due to the low cost and wide application of vision sensors,vision solutions in advanced driving assistant system have been favored by most people.In addition,with the rapid development of machine learning and image processing technology,vision sensors have become more and more important in the field of automotive active safety.The use of monocular vision sensors to realize the functions of lane departure warning and front collision warning has research value.For the lane departure warning system,this paper mainly researches the solution based on gray image,including the image processing method in the system and the implementation principle of the system.The image processing technologies used in the lane departure warning system mainly include image graying,image filtering technology,threshold segmentation and edge detection technology;this article uses Open CV to realize the functions of the lane departure warning system,compares and analyzes the advantages and disadvantages of multiple image processing algorithms in the system,and makes appropriate improvements to some algorithms.In the front collision warning system,the two most important parts are the vehicle identification and the calculation of the distance between two vehicles.In terms of vehicle recognition,this paper selects machine learning algorithms with low hardware requirements for research.There are two main algorithms: Histogram of Oriented Gridients + Support Vector Machines,Haar-like feature + Adaboost cascade classifier.The traditional calculation of the distance between vehicles is mainly done by radar or binocular camera.The solution only using monocular camera can greatly reduce the cost of the system,and this solution has broad development space and application prospect.This paper studies a variety of algorithms that use monocular camera to calculate vehicle distances,and uses the combination of mathematics modeling and camera calibration to calculate distances,which has guiding significance for the improvement of the front collision warning system.