Sonar Sensor Interpretation and Infrared Image Fusion for Mobile Robotics

摘要

We have developed a set of automatic algorithms that use sonar backscattering data to distinguish extended objects in the campus environment by taking a sequence of scans of each object, plotting the corresponding backscatter vs. scan angle, extracting abstract feature vectors and then categorizing them in various phase spaces. We have chosen to perform the analysis with multiple scans per object as a balance between data processing requirements and robustness of the results. Although our current robotic scanner is parked for each scan and then moves to the next scan location before scanning again, it is not difficult to envision a similar mobile robotic platform that scans continuously while moving. It could then take ten or even a hundred scans while approaching a tree or while moving along a unit of a fence, for example. Based on our experience with such scans, however, we would typically expect only the characteristic variations in backscattering behavior described above. Hence, we would envision scans taken continuously as the robot moves towards or along an object, and once the dominant features are identified, the necessary backscatter plots could be processed in the manner described in the previous sections, with the rest of the data safely purged from memory. Our reason for performing this level of detailed processing is a scenario where an autonomous robot is trying to identify particular landmark objects, presumably under low-