Online Motion Segmentation Based on Sparse Subspace Clustering

We consider the problem of online motion segmentation for video streams. Most existing motion segmentation algorithms based on subspace clustering operate in a batch fashion. The main di‐culty of applying these algorithms to real-world applications is that their e‐ciencies can hardly meet the speed requirement when dealing with video streams. In this paper, we propose an online motion segmentation method based on Sparse Subspace Clustering (SSC). Two strategies are adopted in our approach, namely the incremental Principal Component Analysis (PCA) and a warm start from previously obtained Sparse Representation (SR), to accelerate the dimension reduction and SSC in each trail. Through extensive experiments on both synthetic and real-world data sets, we conclude that our algorithm can achieve a signiflcant acceleration under a comparable misclassiflcation rate with respect to other state-of-the-art algorithms.