Abstract
Visual communication or video processing needs video stabilization as a pre-requisite for applications like computer vision or visual tracking. Most of the visual tracking algorithms assume that the input video is stabilized. It is very hard to find any available literature of visual tracking which takes unstabilized video as input for target detection and tracking. It is observed that most of the videos recorded using hand-held camera or camera mounted on a vehicle (tank, ship, aircraft) suffer from unstabilization due to the unwanted hand/vehicle motions coupled with the camera. Target detection is a prime requirement for visual tracking. A novel algorithm is developed which auto-detects the target and uses these parameters to stabilize the video itself. This developed algorithm calculates the local and global motion vectors simultaneously. Local motion vector is used for target detection/tracking while global motion vector is used for video stabilization. Target of interest is identified in frame and different samples of target are taken around the target coordinate. These target samples are then deposited in positive and negative repositories using classifier method. Training of samples of object is carried out for the detection of object coordinate in the next frame. Object coordinate difference between current and just previous frame provides the intentional motion. Consecutive image frames are compensated to get digital video stabilization.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call