Moving Obstacle Detection from Moving Platforms

Developing robust algorithms for moving obstacle detection is a priority for autonomous ground robotic systems. This paper compares various techniques used for detecting moving objects from static and moving platforms and introduces two novel LADAR-based approaches for solving the problem of moving obstacle detection. Video based techniques including: codebook background subtraction, feature tracking and optical flow, and structure from motion are computationally more expensive than LADAR-based approaches and more susceptible to environmental factors, such as level of brightness in the image, as well as clutter and occlusion in scenes. In addition, LADAR-based techniques outperform video based ones when the objects to be detected are visually indistinguishable from the background, which is the case in many battlefield operations. Furthermore, most video based approaches, except for structure from motion, are unable to recover information about the moving targets or obstacles and make numerous assumptions about the input image to achieve their goals. The two novel LADAR based approaches presented, one image based and the other map based, are computationally inexpensive, provide low false alarm rates, and have the ability to easily recover various characteristics of the moving objects, such as speed and direction of travel as well as height and volume information of the obstacles.