A Semantic Based Surveillance Video Indexing and Retrieval System

Beside robust tracking algorithms, a set of efficient and effective tools for video analysis is an important part of security video surveillance systems. The analysis of pre-recorded videos is necessary to discover trends and patterns of threat activities, to evaluate the performance of the security system, and to develop and test new protection rules. The video analysis software should provide security professionals with powerful tools to search and retrieve sophisticated information and knowledge about monitored objects and activities. Currently available content-based video retrieval systems allow for queries on descriptive information, such as target color, location and motion speed/direction, but fall short of searching for high level semantic concepts. Moreover, they lack flexibility as the user has no opportunity to tune the system to achieve best result for a particular query. The objective of this project is the feasibility of an innovative Semantic-Based Video Indexing and Retrieval system (SBVIR) to help security professionals rapidly and effectively search for intelligent information from stored surveillance videos. Our technology is based on a set of powerful statistical classification algorithms that create highly semantic metadata for objects and activities. The system interacts with the user to iteratively increase accuracy and relevance of the retrieved data.