Automatic Detection About Lane And Traffic Light In Intelligent Driver Assistance System

In order to improve the road safety,avoid traffic accidents,this article carried on the research on intelligent auxiliary driving system about lane detection and traffic light detection.Hope can offer some auxiliary information or warning to driver about road conditions before accident with the help of computer vision methods.In the lane detection,first of all,through a combination of adaptive global threshold segmentation,morphology processing and edge detection,this paper achieves stable feature points extraction in different environments;Secondly,this paper proposes a model of the lane.According to the characteristics of the model,the curve parameter can be solved step by step.In the process of solving,we first segment the image according to the distance from the vanish point,the Hough transform and lane reliability calculation based on the Bayesian formula can be used to get the line parameter in near field.Then,according to the tangent of straight line in near field,feature points in far field can be searched by forward method,get the complete curve model;Finally,with the method of particle filter in zonal interested area to track the state of the model.In the location and recognition of traffic lights,first of all,this paper builds the integral channel features of the original image;Secondly this paper proposes a traffic light detection method based on the prior map,using a location priori information and size prior information,the sliding box's threshold is adaptive adjusted for the soft cascade Adaboost,results to improve the accuracy traffic light location accuracy.Finally for areas identified as traffic lights,using Adaboost.MH multi-target classification algorithm for recognition.The innovation of this article is mainly has the following several aspects:First,this paper proposes a innovative approach about lane's feature points extraction method based on morphology,and Canny edge detection and Otsu threshold segmentation results,get feature point in lane and remove disturb at the same time,improve the accuracy of the feature point extraction.Second,in terms of lane line model,this paper proposes a filter method based on Bayesian formula,improves the robustness of the lane line detection in near field;Secondly puts forward a feature point search method in far field,realize the curve lane detection;Finally this paper proposes a line tracking method based on particle filter.Third,this paper proposes an adaptive threshold setting method based on the prior probability map.This method can change the threshold of soft cascade Adaboost classifier according to the position of the traffic lights,improve the accuracy of location.This article has evaluated the performance of proposed methods in a series of experiments,the results show the algorithm is effective in the lane detection and location and recognition of traffic lights.Hoping that that this paper can provide a reference for the related researchers in the field...