Better road following by integrating omni-view images and neural nets

We present results of integrating omni-directional view image analysis and a set of adaptive backpropagation networks to understanding the outdoor road scene by a mobile robot. The road is classified before orientation estimation so that the system can deal with road images with different types effectively and efficiently. A modified omni-view image sensor captures images with 360 degree view around the robot in real-time. New rotation-invariant image features are extracted by a series of image transformations, and serve as the inputs of a road classification network. Then the classification result (the road category) activates one of the road orientation networks to estimate the road orientation of the input image classified in that category. Experimental results with real scene images and some practical problems for actual applications are discussed.