Automatic generic Region-Of-Interest selection for video surveillance applications
Solmaz Javanbakhti, X Xinfeng Bao, Sveta Zinger · TU/e Research Portal · 2014
The value of using cameras in surveillance is further augmented when the surveillance system can take decisions autonomously by means of video analysis.For dynamic scene analysis, automatic detection of informative regions such as Regions-Of-Interest (ROI) is a challenging task for surveillance applications due to the large variations of the scene material.Our hypothesis is that if such an ROI is detected, a further advanced video analysis can be applied later exclusively to that ROI.In this paper, we employ a DCT for an ROI detection, since it provides a compact representation of the signal energy and the computation can be implemented at low cost [1].We verify the usefulness of our hypothesis by cascading our ROI detection techniques with a typical object detector [3] as used in surveillance cases, to evaluate the attractively of the concept.We validate this approach on two different datasets and also compare our algorithm with a number of simple, fast ROI detection techniques.The experimental results show that our proposed approach outperforms the other methods in recall, precision, as well as in computational time.