Feature Extraction Learning for Stereovision Based Robot Navigation System

Stereovision based systems represent the real-world information in the form of a gray scale image known as depth-map with intensity of each pixel representing the distance of that pixel from the cameras. For static indoor environment where the surface is smooth, the ground information remains constant and can be removed to locate and identify the boundaries of the obstacles of interest in a better way. This paper proposes a novel approach for ground surface removal using a trained multilayer neural network and a novel object-clustering algorithm to reconstruct the objects of interest from the depth-map generated by the stereovision algorithm. Histogram analysis and the object reconstruction algorithm are used to test the results.