Abstract: This paper proposes a new algorithm, multi-task robust principal component analysis (MTRPCA),to collaboratively integrate multiple visual features for motion segmentation. Given the video data described by multiple features, the motion parts are obtained by jointly decomposing multiple feature matrices into pairs of low-rank and sparse matrices. The inference process is formulated as a convex optimization problem that minimizes a constrained combination of nuclear norm and `2,1-norm. The convex optimization problem can be solved efficiently with an augmented Lagrange multiplier (ALM) method. Compared with previous methods based on individual features, the proposed method seamlessly integrates multiple features within a single inference
step, and thus produces more accurate and reliable results. Experiments on human motion data sets, Human Eva and change detection, show that the proposed MTRPCA is effective and promising.

Key words:  motion segmentation, low-rank matrix recovery, sparse representation, robust principal component analysis, augmented Lagrange multipliers