Negative obstacle detection for wearable assistive devices for visually impaired

The research on developing assistive devices for visually impaired people greatly relies on image processing techniques. Extensive effort is made on detecting the obstacles in front of the user. Beside the normal obstacles that the user can bump into, another category of obstacles, negative obstacles must also be detected. Negative obstacles are usually represented by holes in the ground or regions that lay bellow the ground surface. Such obstacles represent a potential danger to the visually impaired user and could lead to serious injury if not detected. In this paper, we introduce a stereo vision system to identify and track negative obstacles located in front of the user. The identification is performed in the disparity image based on an estimation of the ground surface in the stereo images. The tracking relies on a camera motion estimation approach. The algorithm we introduce in this paper was integrated and tested with a wearable assistive device to prove its efficiency. We evaluate the accuracy of our solution using various real-life scenarios.