Human detection for a robot tractor using omni-directional stereo vision

It is critical to detect and identify obstacles for safe operation of robot tractors. This study focused on human detection using an omni-directional stereo vision (OSV). The Lucas–Kanade optical flow detection method was used to detect human in a panoramic image. A 3D panoramic image that was reconstructed from stereo rectified images using the sum of squared differences (SSDs) method was used to locate the position of a human. To evaluate the performance of the developed human detection method, two RTK-GPSs were used to investigate the accuracy of the detection method under stationary and motion conditions of a robot tractor. The results of field experiments indicated that a human could be detected successfully under both given conditions in the daytime. The RMS error of measured distance was less than half a meter compared with the reference distance measured by the RTK-GPSs.

► Optical flow detection method was utilized for human detection.
► The distance from a human to a robot tractor was calculated from the stereo image.
► The detection accuracy was evaluated by two RTK-GPSs.
► The RMS error of the accuracy was less than half a meter in the experiment.