Bus Recognition Via Image Processing

Recent years, Chinese economic developed rapidly, at the same time, city’s bombingdevelopment and increasing of vehicles makes the continuous deterioration of the road trafficenvironment. Traffic congestion, road congestion, traffic accidents causes people’s work andtravel has become much more difficult. In order to improve the urban traffic environment andurban road utilize efficiency, government devote a lot to develop urban public transportconstruction, and put efforts to implement bus priority and bus lanes. However, the results arenot promising. The main reason could conclude as poor management methods to avoid illegaloccupation of bus lanes, and solely relying on the inspection of the police. Based on this kindof rural methods, the effect always been limited, in another hand, it is difficult to meet therequirements of real-time management. Therefore, non-bus vehicles’ automatic recognitionbecomes one of the most effective ways to strengthen the management method.In this paper, the background subtraction algorithm is used to get the image of thevehicles from traffic images. For fixed cameras, image’s background has relatively smallchanges. Based on this situation, the methodologies of background subtractionalgorithm could decrease the complexity and get the good results. Histograms of OrientedGradients (HOG) for vehicle’s edge detection can extract accurate feature vectors andminimize the impact of lights, and other environmental factors.In order to find out yellow license plates, which mainly used on Chinese buses, HSI(Hue,Saturation, Intensity) color space is an effective way to detect the color of the vehicle’slicense plate color areas, which could be realized by setting the threshold range of the Hcomponents. On the other hand, based on texture frequency spectrum detection, vehicle’scycle distribution, size and area features would be detected easily. Therefore, with thosecharacteristics, the recognition of bus and truck would get relatively accurate results.There are three important features, vehicle’s HOG, the color of license plate, andvehicle’s texture frequency spectrum, construct the “Support Vector Machine” busrecognition system. According to the bus and non-bus detection, in other words, protocol of“Yes” or “Not”, Linear Support Vector Machine has been designed. The advantages of this system provide the ability of recognition under the small samples conditions, meanwhile,overcomes the traditional problem of the neural networks’ local maxima, with over96%recognition accuracy.