A distributed framework for situation awareness on camera networks

ion for spatio-temporal analysis. Specifically, our contribution includes 1) the design of distributed programming abstraction for spatio-temporal analysis, 2) investigation of performance bottleneck for spatio-temporal analysis on large-scale camera networks, and 3) scalable mechanisms that address computation and communication overheads of state update. 4.1 Application Logic of Spatio-temporal Analysis In this section, we explain the general logic of spatio-temporal analysis on camera networks. Spatio-temporal analysis enables an application to answer queries referring to localityand time-dependent information about different occupants. Common examples of spatio-temporal queries include: “Where was person A at time T?” “When did person B leave zone X?” “When and where did person A and B meet for the last time?” “Who moved from zone X to zone Y between time T1 to T2?” To answer these queries, an application has to maintain its state which represents each occupant’s location at different point of time. Figure 8 shows a general application pipeline of spatio-temporal analysis involving four steps: signature detection, event generation, state update, and query handling. Signature detection involves video analytics to detect signatures such as faces. For example, when a person enters a zone, a face detection algorithm reports the person’s face by analyzing video frames from a camera observing the zone. Note that multiple signatures can be reported from a single frame. For each frame, the face detection algorithm finds all image regions containing faces and passes them to the event generation step. Event generation involves generating a probabilistic estimate about the identity of a detected signature. Depending on the application, different algorithms can be used