Over the recent years, video camera networks have become ubiquitous and a part of our everyday environments. Such camera networks are deployed in a wide gamut of applications ranging from surveillance, emergency response, disaster recovery, and health care monitoring to name a few. Increased availability of computational resources at declining costs, rapid advancements in communication technologies and robust multimedia algorithms for analyzing and drawing inferences from video and audio has led to the deployment of large-scale camera networks involving thousands of cameras in many mega cities of the world. The explosion of video data has led to a new class of applications devoted to real-time automated analysis of CCTV images to create useful information from its contents. Video Content Analysis also referred to as Intelligent Video Analytics can be used to detect, analyze, track and classify the behaviors of people and other objects of interest such as vehicles and packages.
The ability to recognize, reason and retrieve information about human activity is critical to many applications. Central to such applications is the ability of identify and track people as unobtrusively as possible, i.e., even without their active cooperation. Such scenarios range from small-scale environments (monitoring a differently abled person living alone) to medium-scale environments (monitoring elderly patients in nursing homes) and even large-scale environments (surveillance at airports, malls and other public places).
In spite of the many advances in video surveillance arena, the ability of a surveillance system to answer spatio-temporal queries based on the identity of the occupants is limited and mostly requires manual intervention. Transforming the raw feeds from video camera networks into a form that is suitable for information retrieval is a challenging problem, spanning many research areas, including video and audio processing, computer vision, spatio-temporal reasoning, probabilistic databases, and programming abstractions to address scalability of camera networks.
This research proposes to integrate people identification and tracking technologies with spatio-temporal reasoning so as to answer queries about the whereabouts of the occupants of the form ‘Where was X last seen’?, 'What is the probability that Y and Z met in the high-security zone between 6 p.m. and 7 p.m?'. The proposed work has applications in in traffic management and analysis of customer behavior, wide-area surveillance using video cameras, as well as in homeland security-related applications.