In this paper, we presented a method for segmenting moving objects using spectral clustering. The method uses the velocity vectors as the input for clustering, which is more robust to accumulated errors, and then applies spectral clustering in all possible subspace dimensions. The final segmentation is selected from the obtained results using a novel clustering error measure.
Our evaluation on the Hopkins 155 database shows that the method is competitive with current state-of-the-art methods, both in terms of overall performance and computational speed. The algorithm has been shown to be robust to different types of scenes and motions present in the Hopkins 155 database, while remaining very efficient in computation time.