Real-time implementation techniques for particle filter-probability hypothesis density filter based on computational structure analysis

In the construction of multi-target tracking system, the Particle Filter-Probability Hypothesis Density (PF-PHD) filter is difficult to meet the real-time requirements of the system due to its large number of particles and complex filtering process. In this paper, we propose a real-time realization technique of PF-PHD based on computational structure analysis, which firstly improves the particle newborn process in the prediction step of PF-PHD, and proposes an observationdriven particle newborn method based on the pre-newborn particle template. Subsequently, the computational structure analysis of PF-PHD is carried out, and the hardware acceleration scheme based on FPGA is designed according to the data flow and control flow in the filtering process to complete the hardware and software division steps. Finally, the system realization scheme is compared with DSP respectively, when the FPGA running frequency is 150 MHz and the DSP running frequency is 1 GHz, the speed of the hardware acceleration module is about 6.81 times of the DSP processing speed, and the system realization speed is 1.66 times of the DSP realization speed.