A Framework for the Design of Event Detectors

Niels Haering Niels da Vitoria Lobo Richard J. Qian M. Ibrahim Sezan University of Central Florida Sharp Labs of America Orlando, FL 32816 Camas, WA 98607 Abstract We propose a three-level algorithm for the design of event detectors. The rst level extracts texture, color and motion features, and detects motion blobs. The mid-level employs a neural network to verify the relevance of the detected motion blobs using the extracted color and texture features. This level also generates shot summaries in terms of intermediate-level descriptors which combine low-level features from the rst level and contain results of mid-level, domain speci c inferences made on the basis of shot features. The shot summaries are then used by an inference process at the third level that captures the spatial and temporal characteristics of events. We show how the framework can be used to detect landing events in videos and hunts in wildlife documentaries.