A Distributed Framework for Spatio-temporal Analysis on Large-scale Camera Networks
MetadataShow full item record
Cameras are becoming ubiquitous. Technological advances and the low cost of such sensors enable deployment of large-scale camera networks in metropolises such as London and New York. Applications including video-based surveillance and emergency response exploit such camera networks to detect anomalies in real time and reduce collateral damage. A well-known technique for detecting such anomalies is spatiotemporal analysis – an inferencing technique employed by domain experts (e.g., vision researchers) to answer spatio-temporal queries. Performing spatio-temporal analysis in real-time for a largescale camera network is challenging. It involves continuously analyzing the images from distributed cameras to detect signatures, generating an event by comparing the detected signature against a database of known signatures, and maintaining a state transition table that show the spatio-temporal evolution of people movement through the distributed spaces. Being inherently distributed, computationally demanding, and dynamic in terms of resource requirements, such applications are well-positioned to exploit smart cameras and cloud computing resources. However, developing such complex distributed applications is a daunting task for domain experts. In this paper, we propose a distributed framework to facilitate the development and deployment of spatio-temporal analysis applications on large-scale camera networks and backend computing resources. The framework requires the domain experts to provide a set of handlers that perform the domain-specific analyses (e.g., signature detection, event generation, and state update). The runtime system invokes these handlers automatically in the distributed environment consisting of smart camera networks and cloud computing resources. We make the following contributions: (a) a distributed programming framework for spatio-temporal analysis, (b) a careful investigation of the computation/communication costs associated with the large-scale spatio-temporal analysis to arrive at the scalable system architecture, (c) automatic resource configuration to cope with the dynamic workload, (d) a detailed performance evaluation of our system with a view to supporting scalability and quality of service.