Vehicle detection algorithm based on horizontal/vertical edges

This paper presents a vehicle detection algorithm for driver assistance system based on embedded vision architecture. Before generating the hypothesis, we propose to delimiter the search area. The Hough transform is used to detect the lines that delimit the search area. By doing so we can reduce the computation time and the false detection rate during the hypothesis generation phase. The search area is determined by the Hough transform and can be tracking in the following frames. The generation of hypothesis is based on horizontal edges, due to bumper, rear window, plate, etc. and detected by a Sobel filter. The verification of the hypothesis consists to find the "U" shape. For that we verify the presence of two vertical edges, left and right, each connected to a horizontal edge. Satisfactory results have been obtained, representing 90% of correct detection in different situations and with a good rate of 11 frame/s. The algorithm can be implemented on embedded DSP platform.