Multi-Target Tracking with GPU-Accelerated Data Association Engine

Multi-Target Tracking (MTT) is a complex problem in data association and sensor data fusion. By considering a sequence of three-time steps and their triplet costs, a superior solution that incorporates maneuvering target behavior is achieved. This paper presents computational advances in a GPU-based data association engine for MTT, including a fast GPU-accelerated Linear Assignment Problem (LAP) solver, reduction in triplet cost overheads, and computational performance studies. The resulting solution is 5.8 times faster without sacrificing accuracy.