Abstract

Although solutions for robust video processing methods, such as compression or segmentation, have been considerably investigated using general-purpose processors (GPPs), these software implementations are too slow to achieve real-time performance due to the computational complexity and memory bandwidth involved in present complex video processing methods. As such, efficient hardware accelerations are inevitable for fast, video systems. The state-of-the-art, field programmable gate arrays (FPGAs) fill the gap between very inflexible, but high performance ASICs and flexible, yet performance-constrained GPPs. Thus, FPGAs are increasingly employed on hardware platforms in many signal and video processing applications. This thesis proposes an FPGA-based architecture that integrates four video processing methods (sorting, noise estimation, object detection, and contour analysis) on one FPGA, which takes a video signal and outputs a, contour filled video sequence along with the corresponding contour chain codes. The proposed architecture aims at segmenting moving objects in video signals. A video object segmentation consists of several steps: pre-processing (e.g., noise estimation), object detection (i.e., separation of objects and background), and contour analysis. The proposed architecture is simulated, synthesized and verified for its functionality, accuracy and performance on an actual hardware platform consisting of a Xilinx Virtex-4 SX35 FPGA. Compared to related work, our architecture obtains orders of magnitude performance improvements utilizing minimal hardware resources and power, and possesses key algorithmic features, which are inherently required in many video processing applications.