Video stabilization using regularity of energy flow

Jitter and shaky movements of camera are primarily responsible for video destabilization. Such movements usually produce an irregularity in the flow vectors of frames. Video stabilization technique aims to regularize the irregularity of flow vectors. In this paper, an energy-based motion smoothing approach is proposed to smooth the flow vectors using energy of frames. Energy regularity of frames assures a stabilized video while their irregularity causes a destabilized video. Flow vector estimation, motion smoothing and motion compensation are the three primary steps needed for video stabilization. Performance of the stabilization technique depends on each of the above steps, and an optimal method is sought to enhance the performance. In the proposed method, we estimate both the translational and affine flow vectors of a frame using the spatio-temporal regularity flow model. This model provides the approximated flow vectors of all pixels in a frame by minimizing its flow energy function. In the proposed approach, we estimate the flow vectors of the feature points of the maximally stable extremal region of each frame rather than all the pixels of a frame. The proposed video stabilization method is compared with existing state of art methods on the basis of inter-frame transform fidelity, correlation coefficient, regularity and energy of frames. The stability results achieved validate the robustness of the proposed algorithm.