Compressive Sampling for energy efficient and loss resilient camera sensor networks

Data loss is inevitable in multi-hop wireless sensor networks. Multiple packet erasures during transmission can necessitate the use of error-control mechanisms for loss recovery. Energy consumption is also very critical in embedded sensor networks. These problems are even more severe in wireless camera sensor networks (WCSNs), owing to the large data size. Compressive Sampling (CS) turns out to be an effective solution on both the issues. The compression obtained through the linear projections allows transmission of lesser bits than the original. The inherent randomness in CS makes the system tolerant to losses without requiring transmission of redundant parity bits. Both these characteristics help us on saving up on energy. However, using conventional CS on embedded WCSNs has some implementation related challenges. WCSNs mainly find applications in surveillance systems. This requires the snapshots to be large enough to encompass a wide field of view; requiring image sizes at least QVGA or more. The processor memory and the recovery time of L1 optimization, needed for CS recovery, are non-linear with respect to the data size; hence large images hinder the applicability of CS in practical cases. In this paper, we address the issues which may affect the practical usability of CS and provide a CS framework suitable for WCSNs. In order to enable the processing of such large images we propose a block-wise sampling approach, which helps to reduce both the memory overhead and the recovery time. For sampling matrices we use binary sparse random matrices instead of dense matrices, so as to reduce the encoding and decoding times and the computational overheads. Moreover, to further reduce the time factor we employ very sparse matrices (row weight equal to one) and show that they still provide good quality images. We have tested our propositions on WCSNs that include Imote2 sensor nodes equipped with IMB400 multimedia boards, on which we have analyzed the loss resilience of our proposed framework and have also provided an estimate of the energy saved.