Automated panning of video devices
Deepak Kumar, Alice Dsouza, Dutta Kruthika, J Madhushree, Mummidi Manasa · 2017
People can identify the relative position of person and person's movement which fall in their field of view (FOV). This is a complex problem for machine since it has to identify a person in the FOV and also track. In this paper, we are proposing a method as a step towards solving this complex tracking problem. Our objective is to rotate the video acquisition device such that it is constantly tracking the people and their movement. Face detection and recognition are at the forefront of face related computer vision research. We select the first frame from the video stream that is being captured from the acquisition device and perform face detection. A region of interest (ROI) is marked around the detected face and is used as a reference in the proposed method. Histogram is computed for this ROI which act as a reference feature vector. We have employed meant shift tracking to track the ROI through the reference feature vector. We have used Gaussian kernel in our method since mean shift method is a kernel based approach for tracking the object of interest. We evaluate the kernel through back projection using the initial histogram as the reference feature vector. We identify the peak in the kernel and use it for the relocation of our ROI. The ROI gets relocated in every frame according to the peak which is calculated from the kernel placed on the previous ROI. Additionally, we place a slightly bigger ROI on the detected face in the first frame and crop the original frame to the size of bigger ROI. Similarly, every consecutive frame is cropped to form a tracked video which has several applications in surveillance and public addressing systems. In many surveillance and public addressing systems, the video recording is performed by placing the cameras at fixed location. The fixed camera location hinders the objective of tracking person movement. If the person moves out of the camera's FOV, then it is totally uncertain about the person.