Real-time unusual event detection in video stream has always been a difficult challenge due to the lack of sufficient training information, volatility of the definitions for both normality and abnormality of events, time constraints, and statistical limitation of the fitness of any parametric models.


We propose a fully unsupervised dynamic sparse coding approach for detecting unusual events in videos based on online sparse reconstructibility of query signals from a learned event dictionary, relying on an intuition that usual events in a video are more likely to be reconstructible from an event dictionary, whereas unusual events are not.

Our algorithm is completely unsupervised, making no prior assumptions of what unusual events may look like and the settings of the cameras. Also, the fact that the dictionary is updated in an online fashion as the algorithm observes more data, avoids any issues with concept drift.

Unusual Event Detection


Online Detection of Unusual Events in Videos via Dynamic Sparse Coding

Winter 2013/2014