Active Zoom Cameras Control System Using Stereo Parallax

yan gao,xiaolin zhang
DOI: https://doi.org/10.1109/CCST.2010.5678735
2010-01-01
Abstract:Many binocular active tracking systems can track the target by vergence. However, the small vergence angle affects the vergence tracking in wide area. Furthermore, the scale variant problem will also disable the vergence tracking. To get a wide area security, we suggest installing a zoom control method to keep the target in the centre of the image instead of relying on vergence. We propose an active zoom camera system combining the monocular eagle vision and binocular eagle vision. The monocular eagle vision can get a wider field of view (FOV) in high resolution for detection. The binocular eagle vision can maintain the stereo parallax for tracking and zooming. This active zoom camera system can perform four kinds of human eye movements, including the first saccade, the second saccade, smooth pursuit and vergence. We set three zoom cameras at different focal lengths as the peripheral camera for the first saccade, the shallow fovea camera for the second saccade and the deep fovea camera for smooth pursuit. After the first saccade, the peripheral camera zooms in to the same focal length as the deep fovea camera to get the binocular eagle vision. When the deep fovea camera catches the target, the stereo parallax can be calculated to control the smooth pursuit and zooming. At the same time, the constant size of the remote target and the depth of the target can keep the tracking robust. During the smooth pursuit tracking, the shallow fovea camera can provide a wide FOV to avoid the smooth pursuit failure. The parallel processing of zoom control and smooth pursuit tracking control is done in real-time. We report the experiment result of tracking a remote pedestrian from 165 to 110 meters using the stereo parallax. Experimental results show the constant stereo parallax can be attained by zoom control. The zoom tracking can keep the target in the centre of the image as the vergence tracking.
What problem does this paper attempt to address?