Robust Active Object Tracking With Multiple Cameras

Object tracking,refers to select the interested object in the first frame,and then track the object in each frame of the sequence,is an important technology in computer vision.It is a very useful technique for security monitor,video surveillance,humanmachine interaction and autonomous driving.Camera active object tracking,aiming to track an interested object and keep it at the center of view automatically with a movable camera.Compared with static camera tracking,it has a broader field of vision and pays more attention to the object.However,such method,using only one camera and ignoring the information from other views,easily misses the object when there exist occlusion and abrupt motion.In this paper,a robust active object tracking method with multiple cameras is proposed.This method enhances the robustness of single camera object tracking by introducing the object information from multiple views.By acquiring the 3D information,the camera can keep moving effectively and wait for the emergence of the object even though the object is out of view or occluded in some cameras.Specifically,in the process of tracking,each camera tracks the object by the image first,then calculates the 3D position of the object according to the camera's pose and the camera which successfully tracks the object,and drives the camera to track the object continuously through the 3D position of the object's center.The results show that the method can pay more attention to the object of interest,that is,keep the object in the center of the image,and reduces the dependence on the tracking accuracy of single-camera active trackers and is robust to occlusion,abrupt motion and background clutter.At the same time,it also extends the tracking time.Extensive experiments verify its accuracy,effectiveness and robustness on real-world scenes.