A real-time implementation of chaotic contour tracing and filling of video objects on reconfigurable hardware

This paper proposes a real-time, robust, scalable and compact field programmable gate array (FPGA) based architecture and its implementation of contour tracing and filling of video objects. Achieving real-time performance on general purpose sequential processors is difficult due to the heavy computational complexity in contour tracing and filling, thus a hardware acceleration is inevitable. Our finding to the existing related work confirms that the proposed architecture is much more feasible, cost effective and features important algorithmic-specific qualities, including deleting dead contour branches and removing noisy contours, which are required in many video processing applications. Moreover, performance analysis shows that our hardware approach achieves an order of magnitude performance improvement over the existing pure software-based implementations. Our implementation attained an optimum processing clock of 156 MHz while utilizing minimal hardware resources and power. The proposed FPGA design was successfully simulated, synthesized and verified for its functionality, accuracy and performance on an actual hardware platform which consists of a frame grabber with a user programmable Xilinx Virtex-4 SX35 FPGA.