Real-time vehicle classification method for multi-lanes roads

For realizability and real-time processing consideration, a novel vehicle classification method is proposed for heavy traffic flow multi-lanes roads, which can classify vehicles into cars, trucks and buses. In order to monitor two lanes, our system uses three cameras which are mounted overhead of the road and look down the road at an angle of about 60 degrees. Two of them focus on the two lanes respectively to capture the vehicle´s close-up for license plate location and recognition, the left one snaps the two lanes´ panorama and the vehicle features are extracted from it. Firstly vehicles are categorized into cars and noncars roughly according to the color of LPR, and noncars are segmented from scenes by the combination of the position mapping function and local searching. Then five features relating to structural regions are put forward for noncars and their extracted process consists of two main steps: the first step is detecting horizontal edges by a hybrid insensitive noise edge detection method based on Sobel operator and colors, the second step is regions mergence according to colors and positions. Lastly noncars are classified into trucks and buses by a fuzzy rules classifier. Experimental results show that the proposed method is not only robust and accurate, but also can realize the real-time processing with low time-consuming.