Robust Stabilised Visual Tracker for Vehicle Tracking

Visual tracking is performed in a stabilised video. If the input video to the tracker algorithm is itself destabilised, incorrect motion vectors will cause a serious drift in tracking. Therefore video stabilisation is must before tracking. A novel algorithm is developed which simultaneously takes care of video stabilisation and target tracking. Target templates in just previous frame are stored in positive and negative repositories followed by Affine mapping. Then optimised affine parameters are used to stabilise the video. Target of interest in the next frame is approximated using linear combinations of previous target templates. Proposed modified L1 minimisation method is used to solve sparse representation of target in the target template subspace. Occlusion problem is minimised using the inherent energy of coefficients. Accurate tracking results have been obtained in destabilised videos.