On-road Obstacle Detection by Comparing Present and Past In-vehicle Camera Images

We propose a method for detecting general obstacles on a road by subtracting present and past in-vehicle camera images. Compared to the existing learningbased methods that could detect only specific obstacles, the proposed method based on image-subtraction could detect any kind of obstacles. To achieve this, the proposed method first realizes a frame-by-frame correspondence between the present and the past in-vehicle camera image sequences, then performs a road surface registration between the corresponded frames. Obstacles are detected by using the difference of the road surface regions. To demonstrate the effectiveness of the proposed method, experiments were conducted using several image sequences captured by an actual in-vehicle camera. The experimental results showed that the proposed method could detect general obstacles accurately at a distance enough to safely avoid them.