Research And Implementation Of Traffic Lights Detection And Recognition Based On Computational Vision

With the gradual popularization of the car,the traffic situation of the road is becoming more and more complicated,which makes the traffic management and equipment gradually move towards the era of the intelligent.As an important part of intelligent transportation system,the technology of traffic signal lamp detection and recognition in real-time plays an immeasurable role in safety and efficiency of city traffic,it is not only an essential part of unmanned and auxiliary driving,but also is an important guarantee for safe driving to the dyschromatopsia crowd and tired driver.Due to the color of traffic signal lamp and the diversity of environment light,the complex background of the city,and the requirement of real-time,resulting in the current detection and recognition algorithms of traffic signal lamp can not well adapt to the real environment.The current traffic signal lamp recognition is mostly confined to the recognition of circular signal lamp which in simple scenes.According to these problems,the main research of this paper is traffic signal lamp which in the complex city environment,put forward a method about the detection and recognition of traffic signal lamp based on the computer vision.This method not only can meet the requirement of the color recognition in the complex scene,but also can recognize the arrow traffic lamp.Based on the full understanding and analysis of traffic lights characteristics,this paper focuses on three parts,including the traffic signal lamp detection,recognition and tracking.Firstly,pretreat the original image of the traffic signal lamp,then use the algorithms of image color segmentation and morphological filter to locate the position of the black backboard of traffic signal lamp;Secondly,convert the image of black backboard candidate area to the YCbCr color space,use Cb channel image and Cr channel image to realize the location of signal lights,and get the color information of traffic lights;Then normolize the candidate region of lights,use Canny edge detection techniques to extract the contour information of candidate regions of signal lamp,through the extraction and analysis of the improved Hu invariant moment features of signal lights outline,build and select the lights feature sample library,to classify and identify the signal lamp candidate regions;Finally,through analysis of the geometric model of traffic signal lamp,build geometric model of circle,arrow and turn signal lights,verify classification recognition results of traffic signal lights,display the final recognition results,and combine tracking algorithms to track recognized traffic signal lamp.By improving the process of detection and tracking recognition algorithm of the traffic signal lamp,to detect and recognize traffic signal lamp in the complex reality environment,the experiment result shows that this method can improve the accuracy of lights detection,reduce the false detection rate and missed detection rate,realize the recognition of arrow signal lights and turn signal lights,and meet the requirements of real-time.