Real-time visual analysis and search algorithms for intelligent video surveillance

The application of video and image processing algorithms to the field of digital video surveillance has been explored in this work. The aim was to discover the most suitable algorithms for analysing video in real-time and searching large video databases rapidly. The algorithms investigated were designed to extract descriptions of image object content, including features such as shape, colour, texture and motion. A novel approach to using a standardised colour-based search algorithm is described in detail. Results are presented for an experiment that uses this algorithm to retrieve a given person from a video surveillance database. The experimental evaluation shows that the algorithms chosen are both fast and accurate. Our results indicate that a significant timesaving can be achieved when searching video surveillance databases using the described technique.