Research On Vehicle Auxiliary System For Traffic Intersection Based On Machine Vision

With the increasing ownership of motor vehicles, China has been plagued with hightraffic accident rates. According to the statistics of the traffic control department, on averageminute, there is one person being injured or killed. At the same time, the number ofaccidents and casualties rates is significantly increasing at intersection. Violating trafficlights by vehicle drivers and pedestrians is one of the main reasons. With the improvementof machine vision technology and image processing technology, the visual recognitiontechnology has been used to offer intuitive help and prompt for driver. Accordingly, safetyof the vehicle driving has been improved, and the research of vehicle auxiliary systemsbased on machine version has been on the rise in the field of automotive active safety. Thispaper studied numbers of key issues of vehicle auxiliary system as following:(1) Detection and recognition of the traffic light. An intersection traffic light iscomposed by digital indicators and direction indicators. Rapid detection and accurateclassification of the indicators is the key to offer timely and correct auxiliary informationfor the driver. According to this demand, the solution of detection and recognition is dividedinto the steps: fast positioning of traffic light, accurate segmentation of indicators, andvariety of classification methods. First, the position of traffic light is acquired through theglobal threshold algorithm based on color information. Then the traffic light is split into sindependent signal indicators by using the projection method. Secondly, indicators aredivided into digital indicators and direction indicators based on their shape characteristics.Finally, different methods are used to deal with the two kinds of indicator. BP neuralnetwork is used to recognize digital indicator and an algorithm which is combined circledetection and histogram is presented to classify direction indicators. Experimental resultsshow the algorithm executes more efficiently with a better accuracy rate.(2) Pedestrian detection at traffic intersection. Pedestrian detection system is generally higher real-time system, so how to improve the segmentation of the interested region, reducethe image detection area is one research focus. In this paper, behavior of pedestrians attraffic intersection is analyzed. Then the active area of pedestrian is found by detecting thezebra crossing. Through this way, the detection area is effectively narrowed. Histograms ofOriented Gradient feature vectors are extracted and the vectors are fed to a linear SVM forpedestrian/non-pedestrian classification. The experiments prove that this method is feasibleand effective.In this paper, the characteristics of the main objectives-traffic lights and pedestrians atintersection are analyzed, solutions of feature extract and classification are proposed, andthe information of traffic light and pedestrian is achieved to get. It is able to provide supportor alert for driving across traffic intersection. Research has a certain value to improve trafficsafety at intersection, and help to completing vehicle auxiliary systems.