Efficient FPGA Implementation of Parameterized Real Time Color Based Object Tracking

Abstract

This paper presents an efficient color based tracking method applied on a sequence of live video frames for use in real time applications such as surveillance, video conferencing, and robot navigation. The proposed integrated system architecture consists of an attached camera that communicates with the FPGA through HDMI interface. The deployed computer vision algorithm in the FPGA can capture video frames at the rate of 60 fps with the large image sizes of up to 1280×1024 pixels. It then identifies the object based on the specified color, removes noise via spatial filtering and calculates the centroid allowing the object to be tracked during motion. The proposed algorithm leverages a reduction method to minimize the FPGA area as well as power consumption by averaging values over a range of several pixels; thus logarithmically reduces the design size. Our implementation is parameterized to be made as accurate or small as an application requires, with minimal error. The proposed tracking system is implemented on a Xilinx ZYNQ-7000 series XC7Z010 FPGA housed on Xilinx Zybo development board. The utilization reports show for a selected reduction rate of 16, 86.5% reduction in Slice LUTs and 81.3% in Slice registers with the maximum error of 1.5% in centroid calculation.