Obstacle detection and cataloging using radiating sensors

An algorithm for processing range images into discrete obstacles is presented here along with a means of storing them and checking possible paths for intersections. Range images are broken into separate regions for analysis using a sobel mask and flood fill algorithm. A least square plane is fitted to each region, saved as a normal vector, and four points determining the largest possible extent of the obstacle are stored along with vectors from one of the stored points to the corners of the obstacle. With these vectors, colitions between stored obstacles and points on a possible path can be determined with between one and three dot products. The algorithm presented is optimized for fast processing speed and generally overestimates the size of observed obstacles.