Moving Object Tracking Onboard a Moving Platform (MOTO-MP)

Mobile robots are a valuable asset to warfighters, keeping them out of harms way while remotely conducting dangerous bomb-disposal and reconnaissance missions. Autonomous and semi-autonomous operation of these robots would provide additional benefit by freeing up the attention of the operator for other tasks such as personal safety. It also increases usability and reduces training time. Thus far, efforts to develop automation have been limited to static environments. There is considerable resistance to autonomous operation in dynamic environments (which feature multiple humans, robots or vehicles) due to the risk of collision and other safety issues. We propose an approach for “Moving Object Tracking Onboard a Moving Platform”, or MOTO-MP. It enables a robot to detect and track multiple moving objects in a scene while in motion. Unlike traditional approaches which struggle to identify moving objects when the entire scene is in motion, MOTO-MP employs a simple but robust approach involving tracking in image space, an epipolar Hough space, and the 3D world space. A particle filter tracks clusters of moving features, rejecting false tracks, and a Kalman filter predicts the trajectory of objects in world space so that a robot’s navigation and path planning system can avoid collisions.