Development of an Obstacle Avoidance System for a Field Robot using a 3D Camera

The aim of this study was to develop an obstacle avoidance system using a 3D camera. A noise filtering algorithm was used to detect corrupted pixels and a background removal method was used to distinguish obstacles from the background. A clustering algorithm based on a 2D histogram back-projected obstacle pixels onto the ground plane for image segmentation and analysis. Experiments showed that this system had a maximum error of 10 cm and an RMS error of 5.6 cm for position accuracy under static conditions and an RMS error of 7.1 cm in distance detection under moving conditions. Field tests showed that the developed system could effectively detect obstacles in real-time.