Real-time image processing for active monitoring of wide areas

Abstract

A real-time active system able to monitor wide areas by processing video sequences acquired by a pan, tilt, and zoom (PTZ) camera is presented. The system is able to compensate background changes due to camera motion, to detect and to maintain the gaze on objects moving in the scene. Two novelties to speed up the performance of the system and to fit the real-time constraints have been introduced. The first speed-up consists on a new feature clustering method studied to reject badly tracked features. Such method is used to compute a simple yet efficient transformation model for image alignment. The second improvement relies on the introduction of a reference map, containing well trackable features, that is used to select features in a fast and reliable way. The map is maintained and updated continuously by introducing new features related to new regions appeared in the current frame. To detect moving objects, the previous and current frame, after compensation, are processed by a change detection method. Finally, a standard Kalman filter is applied to track objects and to determine the pan and tilt angles that the camera has to perform in order to maintain the gaze on the target.