Algorithm and architecture of low computation FSHEXBS motion estimation for wireless video sensor networks

Abstract

This project reviews the energy efficiency of several popular block-matching motion estimation algorithms that can be used in wireless video sensor network applications. Full search motion estimation provides the best image quality but requires high computing power. Therefore, only fast search algorithms are considered for deployment in wireless video sensor networks, which generally operate in remote and battery constrained areas. However, the image quality of fast search algorithms also needs to be considered in the comparison. The main objective of this project is to design a low computation Motion Estimation Module based on the selection of existing Hexagonal-Based Search (HEXBS) algorithm. It is called Fixed Steps Hexagonal-Based Block-Matching Motion Estimation algorithm (FSHEXBS) due to its reference design name. FSHEXBS is suitable to be used in wireless video sensor networks application while power consumption is the key concern for wireless video sensor networks, video quality need to be in acceptable range. Acceptable range means human eyes ball observation is still able to interpret the video frames objects without serious image distortion. Therefore, FSHEXBS is being proposed due to its lower energy consumption capability compared to existing fast search algorithms included Three Steps Search (TSS), Modified Diamond-Square Search (MDSS), Hexagonal-Based Search (HEXBS), Enhanced Hexagonal-Based Search (EHEXBS) and Full Search (FS). In this project, block-matching algorithms are compared by using two criteria that are computation cost/energy consumption, and image quality. Comparison done across QCIF (176x144 pixels), CIF (352x288 pixels), and 4CIF (740x576 pixels) video formats for 10 different benchmark videos. From the experiment results, FSHEXBS gives the best performance that fulfilled the need of wireless video sensor networks video compression application by offering average 21.32% image quality enhancement and 21.37% of computational cost saving compared to HEXBS. The proposed method is able to reduce the computational load by 95.11% and at the same time having almost equal image quality as compared to Full Search algorithm.