Visualization and summarization of Omnidirectional surveillance Video by Spatio-temporal Indexing

点击翻译按钮获取中文摘要

For surveillance video given from an omnidirectional vision sensor, the goal of this research project is to build a new media system that is capable of visualizing the whole contents by coordinating temporal and spatial representation and summarizing the video by collecting significant scenes. For this purpose, we performed the following research topics and have obtained valuable knowledge and insights.1)Development of processing tools for high-vision omnidirectional vision sensorsA high-resolution ominidirectional sensor with a high-vision camera was introduced and real-time video transformation tools, e.g. generation of perspective images, panoramic images and zoom images were implemented. Also a model of an indoor environment where the sensor was located at its center was constructed.2)Event detection and spatio-temporal indexingAlgorithms for detecting the region of a person with background subtraction and for estimating the two-dimensional position of him/her based on coordinates of his/her legs and head were designed. Viewing IN, OUT, MOVE and STAY as events of the person, we realized event based spatio-temporal indexing. It is confirmed that the spatio-temporal indexing is suitable for surveillance applications.3)Construction of a prototype system for visualization and summarization of surveillance videoWe constructed a prototype system by taking advantage of event based spatio-temporal indexing. To make a viewer easy to understand what event has taken place, the system offers the time-line, spatial map, speed graph and video browsing. In addition, the effectiveness of this system was demonstrated by applying it to actual convenience store environments.