Unsupervised Temporal Segmentation of Human Activities in Video