Human motion segmentation based on low-rank representation

We propose a robust and promising algorithm, named Low Rank Representation (LRR), for addressing human motion segmentation. LRR seeks the lowest rank representation among all the data jointly that represent all data vectors as the linear combination of the base in a dictionary. Given the human motion video, each frame can be regarded as an image, which is a representation of a collection of data vectors jointly. In many cases, the background variations are assumed to be low-rank, while the foreground human motion is sparse. The human motion part can be obtained by removing the low-rank part from the original image. Then the problem is converted to seek the Low Rank Representation of the image. This process is formulated as a convex optimization problem that minimizes a constrained combination of nuclear norm and ℓ2, 1-norm, which can be solved efficiently with Augmented Lagrange Multiplier (ALM) method. Compared to several methods for Human motion segmentation, the proposed method produces more reliable results, yet being more robust to noise and outliers. We do some experiments on the HumanEva human motion dataset. The results show that human motion segmentation by the proposed method is robust and promising.