Power-Rate-Distortion Optimized Resource Allocation for Low-Complexity Multiview Distributed Video Coding

Wireless visual sensor networks are potentially applicable for several emerging applications. Since the data size of the video captured from multiple sensors increases in proportion to the number of video sensors, the efficient compression of video data from multiple sensors is important and still challenging. However, most current multiview video coding approaches extended from single-view video coding standards perform both interview and temporal predictions at the encoder with very high computational complexity, which is not suitable for resource-limited video sensors. In this paper, a resource-scalable low-complexity multiview distributed video coding scheme is proposed. We study efficient exploitation of interview correlation by exchanging the media hash data extracted from video frames of adjacent video sensor nodes at the encoder and using the global motion parameters estimated and fed back from the decoder to improve coding efficiency. In addition, we present a power-rate-distortion (PRD) model to characterize the relationship between the available resources (e.g., power supply and target bit rate) and the RD performance. More specifically, an RD function in terms of the percentages for different coding modes of blocks and the target bit rate under the available resource constraints is derived for optimal block coding mode decision. Analytic results are provided to verify the resource scalability and accuracy of the proposed PRD model, which can provide a theoretical guideline for performance optimization in low-complexity video coding under limited resource constraints. The coding efficiency of the proposed low-complexity video codec is demonstrated via simulation results to outperform three known low-complexity video codecs, especially at high power and low bit rates.