Feature Points Based Video Object Tracking for Dynamic Scenes and Its FPGA System Prototyping
Yin Tsung Hwang, Bing Cheng Tsai, Yu Ting Pai, Ming Hwa Sheu · 2014
A feature points based video object tracking system for dynamic scenes is developed and implemented on a FPGA prototyping platform. In terms of algorithm, based on gradient vectors, the special features of the target object are first extracted and used as a matching template for the object tracking purpose. No object segmentation or background registration is required so that the computing complexity can be largely reduced. The developed algorithm is then mapped to a highly parallel and deeply pipelined systolic array using a space-time transform procedure. Based on the concept of hardware-software (HW-SW) code sign, the entire system is implemented on an FPGA prototyping platform under the control of a host PC. The host PC is in charge of the video capture, display and user interface. The FPGA serves as a hardware accelerator where a hardwired version of the tracking algorithm is realized. A HW-SW communication interface is developed to support the data exchange and control synchronization between PC and FPGA. The experimental result shows that the prototyping system can perform the tracking at a rate of 30fps for 640×480 sized video.