Automatic detection and tracking the moving objects observed by an unmanned aerial vehicles video camera

Abstract

An algorithm of automatic detection and tracking the moving objects for the use in equipment on board of unmanned aerial vehicles is considered. The developed algorithm is based on a tracking specially selected points for a certain period. Tracked points are selected from the areas on the current frame, where the pixel intensity differs from the intensities of the same pixels in previous frames, aligned with the current frame using projective transformation. If the displacement of the tracked points is not fixed on several adjacent frames, they are being deleted, and new points from the areas presumably belonging to moving objects in the current frame are added instead. On each frame the points similar by the location and shape of trajectories of movement are combined into groups that presumably correspond to moving objects. Objects are tracked by comparing the groups of moving points with the points of neighboring frames. Groups of moving points from neighboring frames are matched if they contain a large number of common tracked points. The algorithm allows simultaneous tracking of more than 20 objects in real time. The indication of objects as moving occurs only if during the time of its tracking it has shifted a considerable distance. The algorithm has a low percentage of false detections of moving objects, it detects well small objects and is capable reliably to accompany moving objects.

Keywords

• motion detection
• objects tracking
• unmanned aerial vehicles
• moving camera
• optical flow