A Real-time Vision-based Outdoor Navigation System for the Wheelchair Robot

This paper presents a robust and reliable system that navigates the wheelchair robot without manual initialization or a priori information in real-time in outdoor environments. The system can extract various lane boundaries including straight and curved lanes under different conditions. It is robust against noise, shadows, and illumination variations in the captured road images. The straight lane detection algorithm and curved lane detection algorithm can automatically function in the correct mode based on the current road type. The straight lane detection algorithm uses three features to represent two lane boundaries. The curved lanes are segmented using the lane-curve function that is generated by transforming a defined parabolic function in a world coordinate into an image coordinate. The accuracy of the navigation system is 100% after the space-search based correction.