Bayesian algorithms for adaptive change detection in image sequences using Markov random fields

Background Subtraction on Embedded Hardware

“…Typical Markovian models are based on a maximum a posteriori formulation that is solved through an optimization algorithm such as iterative optimization scheme (ICM) or graphcut [2,112] which are typically slow. In [14] it was shown that simple Markovian methods (typically those using the Ising prior) produce similar results than simple post-processing filters.…”

Change detection and background extraction by linear algebra

“…Since they are functions of the illuminance as mentioned in Section V-C, the derivative is simply . Example 6.1: The linear combination of these functions is (21) The Wronskian for this equation is then (22) Example 6.2: Solving (21) for yields (23) and the Wronskian is (24) where .…”

Transformation of adaptive thresholds by significance invariance for change detection

“…in terms of Gibbs/Markov random fields (GMRF) [26,27]. Spatial GMRFs are suitable to express the prior expectation of compactly shaped moving objects, and consequently suppress the emergence of spurious, noise-like detection results [16].…”

“…the rates of false positives and false negatives. Moreover, since motion detection as a binary segmentation problem seeks to infer the underlying structure from observed noisy image data, we added it [16,17] to the list of ill-posed problems in computer vision [18,19,20]. When solving such problems, the space of possible solutions is constrained by appropriate regularization [21,22,23,24], which may also be expressed in a statistical manner within Bayesian approaches [25,7].…”