Adaptive search window for low complexity surveillance video coding

Abstract

Static video surveillance is an effective tool for maintaining security. Videos from stationary cameras contain mainly background, with movement appearing rarely. Standard video coders record everything in the video frames at the same quality, including the background which contains no information of interest. Knowledge of the application could improve the video compression efficiency. Standard video encoders takes a lot of computation power, with Motion Estimation (ME) taking around 60-80% of the encoder compute budget. Encoder computation cost can be reduced by encoding the background region at lower quality. This paper proposes the use of adaptive search window to reduce the search range to achieve the objective of lowering computation complexity.