Multiscale edge fusion for vehicle detection based on difference of Gaussian

Edge information help highlight the contour as well as cast shadow of objects. As the low complexity for edge extraction, the edge-based methods are widely used in vehicle detection. Traditional edge-based vehicle detection methods are easily interfered by noise and background, which resulting in inaccurate false detection. In this paper, a vehicle detection method based on multiscale edge fusion is proposed. First, multiscale images are obtained from the decomposition of the DoG pyramid. Second, multiscale edges are extracted by the DoG operator in multiscale images. Third, different scale edge map are fused according to the proposed multiscale edge fusion strategy. Then, an accurately located, low redundant and strongly anti-noise edge map is obtained. Finally, morphological operation and connectivity analysis are applied on the edge fusion map. Experiments with traffic images in different weather conditions verify the practicability of the proposed method. Comparison with related method in detection rate and detection accuracy verifies the superiority of the proposed method.