Compressive Sensing for background subtraction based on error correction coding

Compressive Sensing (CS) provides a solid theoretical framework for sparse signal recovery. In this work, we concentrate on recovering the foreground object which can be represented as a sparse vector using wavelets. The method revolves around the CS framework by judiciously using the complex field BCH codes and the syndrome as measurements to achieve our goal of robust background subtraction using reduced number of measurements. We propose static as well as adaptive techniques for acquiring the measurements required for recovering the images. The reconstruction, which is carried out by a Complex-field BCH decoder coupled with block-based implementation provides elegant computational advantage compared to the conventional CS reconstruction methods. The proposed techniques have been experimentally validated with surveillance video sequences in a thorough manner and the results are summarized.