A driver assistance system typically adopts the wide-angle camera to obtain a wide-view image. However, the wide-angle camera often produces radial distortion. Since the conventional training-based pedestrian detection method uses distortion-free training samples, it is not suitable for distorted images. In this paper, we propose an effective pedestrian detection method that divides pedestrian training samples into several classes according to the amount of radial distortion, and trains each class separately. Likewise, a test image is divided into sub-regions and detection is performed for each sub-region separately. Experimental results show that our approach provides better performance compared to the conventional method.