Parallel multitarget tracking particle filters using Graphics Processing Unit

This paper presents design, implementation, and performance evaluation results of parallel multitarget tracking particle filters using a Graphics Processing Unit (GPU) as a parallel computing environment. Since the resampling step is a bottleneck in parallel particle filtering two parallel tracking algorithms are implemented and compared - with centralized resampling and with distributed resampling using proportional allocation, respectively. Performance evaluation and comparison between both parallel implementations, and with a fully centralized implementation are provided based on experimental data obtained from several simulated multitarget tracking scenarios.