Studies on Remote Surveillance by Cameras

点击翻译按钮获取中文摘要

Unceasing surveillance are sometimes needed on sites like nuclear power plants, transformer substations, or parking areas. Although the job is now normally done by humans monitoring views from surveillance cameras, the monotonous work causes physical fatigue and mental stress. In this research, basic studies on surveillance aiding systems are done. The following are the studies :1. We developed methods for detecting moving regions, applying background/interframe subtraction method even while the background of the view is moving by rotation, of pan-tilt cameras. To do this,we constructed projection image to virtual cylinder screen, applied template-matching for multiple templates, estimating translation of background by voting, and compensate the background translation.2. We proposed new methods for efficiently recording and searching spatio-temporal events using data structure called PMD-tree. Test results of the performance was also presented.3. Displaying sites under surveillance by 3D CG can be helpful for observers. We proposed methods for constructing 3D virtual environment from several photos, and constructing 3D virtual deformable object model from large number of views.4. To learn and classify spatio-temporal event patterns, we proposed graph representation that includes layout information and developed algorithm for extracting frequent patterns.