Off-highway obstacle detection

The biggest conclusion is that intensity of radar backscatter returns is a poor indicator of danger to a vehicle despite the attempts of several projects to show otherwise. Still, our method violates this finding at some level. An image of backscatter intensities is filtered with various image processing techniques then is thresholded as if it has become an image of risks or confidences. Current research is investigating discriminant functions that allow arbitrary numbers of classes and can use separability and discriminability as confidences instead of a filtered version of intensity. It cannot be overemphasized that this system was designed as an add-on sensor to an already operable autonomous vehicle. As such, it is not sufficient as a primary or stand-alone sensor in an unstructured environment. Radar techniques to detect road edges [Kaliyaperumal et al., 2001], [Nikolova and Hero, 2000] or terrain quality would fill these gaps and allow a radar-only, all-weather autonomous platform. Autonomous radar navigation is possible with proper scoping of the problem. On highways this means ignoring cross-sections smaller than a motorcyclist. Off highway the solutions must be a little more creative, but still exist. Applying image processing techniques already well developed by the computer vision field and avoiding the assumption that intensity of backscatter is a feature directly related to obstacle danger will help advance this goal. Small radar systems have recently become very good, but they do not solve the problem alone. Just as other sensors like cameras and LIDAR require algorithms to convert from raw data to obstacle classifications, so too should radar users not expect to achieve good performance by simply thresholding the output from hardware. If radar is seen as just another imaging source like a camera, much work can be accomplished using the existing arsenal of image processing