The MS-Detect software written in C++ enables to detect moving objects in image sequences by background subtraction. It simultaneously provides at each time instant the binary map of mobile objects in the scene and the reconstructed image of the static background.
The key point of our joint approach is to define a single random process that can take two types of value (instead of defining two different processes): one symbolic (for motion detection) and one numeric (for background intensity estimation). Thus, we can properly handle space-time interactions between moving points and background information. It is expressed by a mixed-state conditional random field. The solution is obtained by minimizing the resulting energy function. Experiments on real sequences and comparisons with existing motion detection methods support our proposal. The algorithm also provides with video sequence inpainting (filling the moving regions with background).
You can try MS-Detect on our mobyle web portal: Try MS-Detect
Left: Fluorescence microscopy image sequence of Rab6-GFP proteins in a micro-patterned circular-shaped cell acquired with spinning-disk (MIP) (from UMR 144 CNRS – Institut Curie).
Right: Detected moving components plotted over the original images.
Left: original video sequence (static camera)
Middle: sequence of the estimated mixed-state fields, in white the detected moving regions
Right: reconstructed image sequence of the static background (the moving parts are progressively filled in,
T. Crivelli, P. Bouthemy, B. Cernuschi-Frias and J.-F. Yao. Simultaneous motion detection and background reconstruction with a conditional mixed-state Markov random field. International Journal of Computer Vision, 94(3): 295-316, 2011.