Graph based event detection from realistic videos using weak feature correspondence.

We study the problem of event detection from realistic videos with repetitive sequential human activities. Despite the large body of work on event detection and recognition, very few have addressed low-quality videos captured from realistic en- vironments. Our framework is based on solving the shortest path on a temporal-event graph constructed from the video content. Graph vertices correspond to detected event primi- tives, and edge weights are set according to generic knowl- edge of the event patterns and the discrepancy between event primitives based on a greedy matching of their visual features. Experimental results on videos collected from a retail envi- ronment validate the usefulness of the proposed approach.