Object Trajectory Association Rules for Tracking Trailer Boat in Low-frame-rate Videos

Tracking object accurately in one frame per minute (1-fpm) video is believed to be impossible, because the one-minute discontinuity of object coupled with dynamic background variation implies that the motion and appearance of target is theoretically not predictable. In the context of maritime boat ramps traffic surveillance, we propose in this paper a novel approach to tracking object in the low-frame-rate (LFR) of 1-fpm videos, where the motion discontinuity of object is mitigated by adopting target lifespan path-template and association rules of behavior prediction. The approach has been applied to trailer boat counting at three maritime boat ramps in New Zealand. The obtained accuracy goes above 90 %, with reference to the ground truth manual counting.