Abstract
We present a complete end-to-end framework to detect and exploit entry and exit regions in video using behavioral models of object trajectories. Using easily collected âweakâ tracking data (short and frequently broken tracks) as input, we construct a set of entity tracks to provide more reliable entry and exit observations. These observations are then clustered to produce a set of potential entry and exit regions within the scene, and a behavior-based reliability metric is used to score each region and select the final zones. We also present an extension of our fixed-view approach to detect entry and exit regions within the entire viewspace of a panâtiltâzoom camera. We additionally provide methods employing the regions to learn scene occlusions and causal relationships from entryâexit pairs along with exploitation algorithms (e.g., anomaly detection). Qualitative and quantitative experiments are presented using multiple outdoor surveillance cameras and demonstrate the reliability and usefulness of our approach.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have