Motion estimation based frame rate conversion hardware designs

Official URL: http://192.168.1.20/record=b1307454 (Table of Contents)

Frame Rate Up-Conversion (FRC) is the conversion of a lower frame rate video signal to a higher frame rate video signal. FRC algorithms using Motion Estimation (ME) obtain better quality results. Among the block matching ME algorithms, Full Search (FS) achieves the best performance since it searches all search locations in a given search range. However, its computational complexity, especially for the recently available High Definition (HD) video formats, is very high. Therefore, in this thesis, we proposed new ME algorithms for real-time processing of HD video and designed efficient hardware architectures for implementing these ME algorithms. These algorithms perform very close to FS by searching much fewer search locations than FS algorithm. We implemented the proposed hardware architectures in VHDL and mapped them to a Xilinx FPGA. ME for FRC requires finding the true motion among consecutive frames. In order to find the true motion, Vector Median Filter (VMF) is used to smooth the motion vector field obtained by block matching ME. However, VMFs are difficult to implement in real-time due to their high computational complexity. Therefore, in this thesis, we proposed several techniques to reduce the computational complexity of VMFs by using data reuse methodology and by exploiting the spatial correlations in the vector field. In addition, we designed an efficient VMF hardware including the computation reduction techniques exploiting the spatial correlations in the motion vector field. We implemented the proposed hardware architecture in Verilog and mapped it to a Xilinx FPGA. ME based FRC requires interpolation of frames using the motion vectors found by ME. Frame interpolation algorithms also have high computational complexity. Therefore, in this thesis, we proposed a low cost hardware architecture for real-time implementation of frame interpolation algorithms. The proposed hardware architecture is reconfigurable and it allows adaptive selection of frame interpolation algorithms for each Macroblock. We implemented the proposed hardware architecture in VHDL and mapped it to a low cost Xilinx FPGA.