Online Learning for Video Probabilistic Appearance Manifolds Recognition Algorithm and its Application in Coal Mines

Abstract

To monitor the operation of coal mine safety production, an online learning method of fault identification for coal mine safety production through video probabilistic appearance manifolds is proposed in this paper. For a category of the coal mine equipment safety state, a common representation of the normal appearances of this category would usually be learned off-line. From video monitoring of this category, an appearance model can be learned online through a prior generic model and successive video. The further details, as well as both the normal and abnormal appearances, can be expressed as an appearance manifold. In our algorithm, an appearance manifold would be approximately estimated by a series of sub-manifolds, and each sub-manifold is further refined into a low-dimensional linear sub-space. Thus, the time required for image recognition is reduced to meet the demands of real-time image processing. Through experimental analysis, we can demonstrate that our online learning algorithm method is an efficient method for video-based image recognition, and its application in coal mine safety production has proven to be very effective.