High Level Synthesis Implementation of Object Tracking Algorithm on Reconfigurable Hardware

Abstract

This paper presents a variation of Hough Transform (HT) algorithm implementation for symmetrically shaped object detection in the real-time video systems. The proposed algorithm is implemented on Xilinx ZYNQ-7000 series XC7Z010 SoC using High-Level Synthesis (HLS). The integrated system architecture contains a high definition camera that communicates with the SoC through HDMI interface. The algorithm is efficient in terms of memory space, computational resources, performance and accuracy as it only uses one-dimensional HT space instead of two-dimensional HT space. The proposed algorithm does not require an external memory for storing the HT space and does not require complex computations to detect the object. The experimental results show that only 40 18K BRAMs are used to implement the proposed algorithm at 60 frame per second. The proposed architecture is implemented for the video frame size of 1920 × 1080 pixels, at maximum latency of approximately 2099032 clock cycles, i.e., 14.69 ms for this HD video frame. Our algorithm uses only 26 DSP blocks for radius calculation while the entire video processing design uses 19150 slices of ZYBO SoC.