Active Stereo Visual Sensing with Dynamic View Planning for Enhanced Activity Observation
DescriptionMost of today’s surveillance systems use 2D images from fixed cameras, which cannot perform adequately in many situations. In this project, an active stereo vision system will be developed to achieve enhanced performance in activity observation and perception for advanced surveillance applications. The proposed system is designed to observe target motions and percept/recognize its activities by dynamically changing the viewing pose and the settings of the sensing system on-line. Methods for automatic recalibration of the vision system will be developed. This project will focus on the target motion trajectory including its extraction, description, analysis in activity perception/recognition. A new activity modeling method will be developed based on trajectory description. The proposed system can plan its next view to best observe the predicted activities of the target, which is useful for many applications of advanced surveillance involving activity recognition/perception such as robot learning by demonstration, human activity monitoring and security control.
|Effective start/end date||1/10/11 → 4/03/16|