In today’s camera-based surveillance solutions video streams are both recorded for later inspection and displayed on-screen for human operators. This requires the attention of operators for unusual events and urgent response. To complicate matters, human operators have limited attention spans as well as cognitive limits on how many different video streams can be observed simultaneously.
Consequently, unusual scenes might not be noticed. Searching for unusual events in large-scale video databases is a tedious, time consuming task for humans.
Therefore the bigger part of the information hidden in these videos stays unused unless systems are able to automatically process them.
The Unusual Scene Detector is an automatic notification system for unusual appearances. In this context, something is unusual if a similar appearance has not been or has rarely been observed in the past on the same location.