온보드 고속 목표물 추적을 위한 FPGA 기반 양자화 병렬 입자 필터링

초록

This paper proposes a hardware-accelerated particle filter architecture for real-time, high-speed target tracking in on-board system, combining clipping-based symmetric quantization with parallel processing. To overcome the bottlenecks of conventional sequential methods, we designed custom IP cores on a Zynq FPGA for Model Propagation and Resampling, optimized with pipelining and unrolling techniques. The proposed architecture achieved a 48% reduction in total execution time compared to a Processing System (PS) implementation. This work validates our design as an effective solution that balances real-time performance and precision, showing strong potential for next-generation AI processors.

키워드