A computational random access memory (RAM) architecture for real-time video motion tracking

Computational Random Access Memory (C-RAM) has proven to be a candidate solution to carry out massive parallel computations. It benefits from the large internal memory bandwidth be having processing elements implemented inside the memory. Several C-RAM architectures have been used in the literature to carry out different computational intensive tasks but they failed to achieve real-time operation of the video motion tracking process.; This thesis presents a new C-RAM architecture for video motion tracking using a novel mesh-based motion tracking method. The proposed motion tracking method overperforms the commonly used motion tracking methods in terms of peak signal to noise ratio (PSNR). In the proposed C-RAM architecture a large number of processing elements has been embedded with SRAM blocks to utilize the large internal memory bandwidth. All the processing elements are working as a single instruction multiple data (SIMD) architecture. Parallel and pipelined implementation has been used to achieve the real-time operation of the proposed architecture. The proposed architecture needs 27.49 ms to process one CIF video frame, which meets the real-time requirements of processing thirty frames per second. The proposed architecture has been prototyped using 0.18 mum CMOS TSMC technology and the embedded SRAMs have been generated using the Virage memory compiler.