A reliable vehicle detection-location framework based on homography planar projection

Considering vehicles in highway as surveillance object, the paper proposes a detection-location framework based on homography planar projection. The proposed work allows vehicle' position, speed value and direction angle in physical scale to be accessed. Different from the traditional detection, our detection part employs background modeling and detector trained by AdaBoost one after another to extract vehicle wheels from the scene. So its results is precise enough. At the same time, the shadow suppression is improved by region analysis on hue, that reduces the shadow's influence on contour extraction so that the adjacent vehicles can be separated well. Another contribution of this paper is that, homography relation between imaging plane and ground is ensured by using some designated symbols in highway, which could be used to compute the real location of the detected vehicle wheel. And some interesting information also can be estimated with this method, such as speed value and direction angle. Finally, some experiments is done with filed image sequences and it shows that the presented method is effective.