Target Tracking System Using Kernelized Correlation Filtering Based on FPGA

Driven by the demand for highly real-time target localization and tracking on size, weight, and power (SWaP)constrained edge platforms, this article proposes a lightweight, all-hardware streaming correlation filter (CF) tracking framework implemented on a field-programmable gate array (FPGA). At the input, this method uses adaptive scaling to achieve a fixed-size computation window. In addition, multiple configuration combinations are evaluated to determine a more suitable setup, balancing accuracy and implementation overhead. The tracking core uses pure logic scheduling to orchestrate initialization, localization, and updates. It minimizes data movement and peak resource usage through intraframe readback buffering and cross-stage time-division multiplexing of the frequency-domain transform unit, ensuring stable throughput. The experimental results show that the prototype achieves a maximum end-to-end latency of 3.92 ms and a power consumption of 1.64 W. Tests with 720p video streams verify stable real-time tracking, and stress tests demonstrate a sustained throughput exceeding 250 FPS. These results indicate that the proposed design provides a simple and efficient solution for high-speed, low-power visual tracking systems.