Background subtraction via incremental maximum margin criterion: a discriminative subspace approach

摘要

Background subtraction is one of the basic low- level operations in video analysis. The aim is to separate static information called "background" from the moving objects called "foreground". The background needs to be modeled and updated over time to allow robust foreground detection. Recently, reconstructive subspace learning models, such as principal component analysis (PCA) have been used to model the background by significantly reducing the data's dimension. This approach is based on the assumption that the main information contained in the training sequence is the background meaning that the foreground has a low contribution. However, this assumption is only verified when the moving objects are either small or far away from the camera. Furthermore, the reconstructive representations strive to be as informative as possible in terms of well approximating the original data. Their objective is mainly to encompass the variability of the training data and so they give more effort to model the background in an unsupervised manner than to precisely classify pixels as foreground or background in the foreground detection. On the other hand, discriminative methods are usually less adapted to the reconstruction of data; although they are spatially and computationally much more efficient and often give better classification results compared with the reconstructive methods. Based on this fact, we propose the use of a discriminative subspace learning model called incremental maximum margin criterion (IMMC). The objective is first to enable a robust supervised initialization of the background and secondly a robust classification of pixels as background or foreground. Furthermore, IMMC also allows us an incremental update of the eigenvectors and eigenvalues. Experimental results on different datasets demonstrate the performance of this proposed approach in the presence of illumination changes.