Dynamic Joint Probabilistic Data Association Framework for Target Tracking with Ground Robots

With the rise in use for ground robots, there is a need for efficient detection and tracking to improve SLAM performed by each such agent. This paper analyses the effectiveness of the Dynamic Joint Probabilistic Data Association (DJPDA) framework for target tracking in dense environments by creating a test bed with a fleet of ground robots. DJPDA is a framework that utilizes tensor decomposition, a commonly used technique to tackle curse of dimensionality, to handle the exponential growth in the binary non-competing joint association events (or feasible events) in Joint Probabilistic Data Association (JPDA) filter. The number of feasible events in JPDA is reduced by utilizing the core tensor, a result of the tensor decomposition, as a surrogate for the input of JPDA instead of the complete set of measurements. The test bed created for this experiment consists of 5 Kobuki ground robots. The laser scan data from the on-board Xbox Kinect sensor is collected for each time step by using one of these robots as the observer. Finally, the collected point cloud data is passed to the DJPDA framework for offline computation of the tracks to compare these predicted tracks with the true tracks obtained from the odometry data for each ground robot.