Time-to-collision (TTC) is an indicator for collision avoidance which takes into account the relative velocity of a subject and an object which concerns the subject. The most common approach for estimating TTC is to employ optical flow. However, this method takes a huge processing time. In this paper, we propose a new estimation algorithm to extract the collision possibility of an approaching object. Then, it is eventually used for collision avoidance. With this new algotithm, TTC can be robustly and real-timely obtained by processing simple information from a moving object. Here, the necessary information are the number of active blocks and the changes in that number, and the changes in the center position of the actives blocks. Using these information, the collision possibility is then estimated by fuzzy inference where if the collision possibility is large, TTC is estimated and used for collision avoidance. The effectiveness of the proposed method is confirmed by several computer simulations employing images with dynamic scenes.