Warped K-Means: An algorithm to cluster sequentially-distributed data
Information Sciences: an International Journal
| Hi-index | 0.00 |
Many problems in Computer Science require a trajectory segmentation, in part due to the notably huge spectrum of devices that capture sequentially-generated information (e.g., motion sensors, video cameras, RFID tags, eye trackers, etc.) Segmentation leads to simplify the structure of the data, so that original objects can be divided into smaller, more compact structures. Seen this way, segmentation can be approached as a compression technique, i.e., organizing trajectories into segments whose members are similar in some way. This can be solved as a clustering problem. Unfortunately, to date we have not found a suitable method that can tap in a really simple way the temporal constraint implicitly embedded in the data. Moreover, near-optimal solutions such as kernel methods or hidden Markov models can be prohibitive if processing power is a restriction (e.g., on mobile devices).