Video-based Research On Bus Vehicle Detection And Identification

Vehicle detection and identification technology is important to intelligent transportation system.Traditional detection method like induction coil and ultrasonic detector has been very difficult to meet the requirements of low cost,high reliability and long service life.This paper researches bus video detection by foreground of moving object,classifier detection and color characteristics methods.Specific contents of this paper are:(1)The method of combination foreground,together with morphological processing and threshold method of removing shadow are used to finding out the moving vehicle from video sequence.The paper researches and analyses common filtering methods like the median filtering,Gauss filtering,bilateral filtering,and choose the better method for bus detection.Comparing foreground extraction methods of frame difference and optical flow with background modeling methods of Single Gauss model,Mixed Gauss mode and mean background model,final Mixed Gauss mode and frame difference are used.(2)The Ada Boost algorithm and Haar feature is used to detect the bus from moving foreground.Analysing Haar and LBP feature,integral map and the process of classifier training,the Ada Boost algorithm with feature of Haar and LBP is used to train two strong classifiers.Comparing the detection results of the two characteristics,also considering the time of each classifier has spent and the real-time requirements of the vehicle detection,then the Haar feature is used.(3)The methods of windows location and the ratio of red pixel number in windows are used to distinguish between bus and passenger vehicles.Comparing bus with passenger vehicles,bus has characteristic color which is to mark lines in both front and rear windows,other passenger vehicles has less than bus.Windows location and the ratio of red pixel number in windows methods are used to detect the result of classifying.Three kinds of edge detection methods like Canny,Sobel and Laplacion combine with morphological and connected domain processing are used to locating the windows,then finding the best method of windows location.Transfering the windows area into HSV color space,counting the ratio of characteristic color pixels,and comparing with threshold that has been set.If the ratio is larger than threshold,then believe it is bus,otherwise it can’t be bus.