Non-recursive method for motion detection from a compressive-sampled video stream

This paper introduces a non-recursive algorithm for motion detection directly from the analysis of compressed samples. The objective of this research is to create an algorithm able to detect, in real-time, the presence of moving objects over a fixed background from a compressive-sampled greyscale video stream. Many difficulties arise using this type of algorithm because it violates the fundamental principles of compressive sensing reconstruction that lie beneath traditional recursive methods. Recursive reconstruction methods even if accurate need large amounts of time and resources because they aim to retrieve all of the information contained within a scene. Our method is based on two key considerations. The first is that the targeted information of a moving element compared to a fixed background is really small. The second is an appropriate choice of a sub-Gaussian compressive sampling strategy. Our aim is to reduce the focus of general reconstruction in order to retrieve only objects of interest. This algorithm can be used to process compressed samples derived from a video stream with a speed of 100fps. This makes possible to detect the presence of moving objects directly from compressed samples with limited resources.