Abstract
Many researches have been done as for outdoor scene recognition by analyzing optical flow information. However, a practical driver-oriented system has not been proposed yet that notices the “Risk degree” when multiple obstacles exist, such as, parking cars and pedestrians. This article presents a new method to detect stand still obstacles using the “model flow” corresponding to optical flow field where there is no obstacle. The model flow is calculated using the knowledge as for navigation environments. Standard Points which are regularly set on image coordinates are transformed to Environment Points on road plane using several parameters such as, steering angle and camera tilt angle. Stand still objects are automatically detected by analyzing the difference between a model flow field and a real optical flow field which is acquired by an in-vehicle camera image. The algorithms have been installed in a desktop PC by Microsoft Visual C++6.0 programs. Experimental results by several kinds of outdoor scenes have shown the effectiveness of the proposed method. Total detection ratio is 91.0% and that in Dangerous case is 100%.
