In a nutshell, the technical objectives of CS-ORION are to employ, implement, and validate the concepts of compressed sensing in the capture, coding, transmission, and reconstruction of image and video for power constrained remote surveillance systems.  Our goal is to consider a long-term, multi-layer approach that combines expertise from statistical signal processing, data representation theory, and video coding and transmission techniques, for enabling robust and high-quality remote imaging.  Our approach is to employ compressed sensing signal acquisition principles and Bayesian reconstruction methods so as to address the need for novel video compression techniques that can achieve good performance with a computationally light encoder, possibly shifting some of the system complexity to the decoder.

Compressed Sensing is a new framework recently introduced for simultaneous sensing and compression and it presents an extension to the Nyquist/Shannon sampling theorem.  According to the CS framework, a signal can be under-sampled without losing information (i.e., without sensitivity loss), provided the signal is sparse in some basis, and that the basis on which the signal is sampled is incoherent with the basis on which the signal is sparse.  More quantitatively, if a signal consisting of N samples is K-sparse, then mathematics theory shows that it can be represented by only O(KlogN) samples.  CS is the projection onto incoherent measurement ensembles and the reconstruction can be achieved by solving a convex optimization problem.  CS enables a potentially significant reduction in the sampling and coding computation costs at a sensing device with limited capabilities.

The formalism of compressed sensing presents a number of key properties that could be extremely advantageous in the application of remote surveillance:

• Fast, simple and efficient compression/coding that is perfectly suitable for a remote imaging system where on-board software has to run with minimal CPU load, memory, and power requirements.
• Fully linear compression that allows very efficient data fusion (i.e., summing different reconstructed datasets to improve the signal to noise), contrary to non-linear compression methods (e.g., JPEG, MPEGx) that nearly prevent data fusion.
• Full decoupling between the compression/coding stage and the decompression/reconstruction stage, implying that this second stage can be continuously improved even if the first stage is fixed, and can also include priors to improve the reconstruction.
• Since the signal is projected on a random set of measurement vectors, what is transmitted is very robust to bit losses that could happen in the case of usually adverse operating remote sensing environments.
• Compressed sensing measurements can be viewed as “weakly encrypted” for an attacker without knowledge of the measurement matrix.  Compressed sensing-based encryption provides both signal compression and encryption guarantees, without the additional computational cost of a separate encryption protocol and it could be useful in video surveillance, where implementing an additional software layer for cryptography could be costly.