Real-Time and Robust Video Stabilization Based on Block-Wised Gradient Features

Portable camera devices are commonly used in modern society. However, the videos captured by such consumer electronics cameras usually exhibit heavy shake which affect the consumer experience. Also, consumer cameras often have limited computational resources. Thus, how to efficiently eliminate camera shake has become a hot issue in the field of consumer electronics. This paper proposed a fast video stabilization algorithm based on block-wised gradient features. After extracting the gradient features, the block-based feature descriptors are generated to perform feature point matching. Then, we calculate the affine transformation matrix of two consecutive frames of the video sequence to register the first frame to generate the final stable video. Experiments show that our fast method has excellent real-time performance and stabilization performance. To deal with complex shaky scenes, we further proposed an improved robust method. In our improved algorithm, the camera's trajectory from the transformation matrix between two frames is extracted and the total camera path is calculated to acquire the final stabilized video. The results of experiments demonstrate that our improved method can perform real-time stabilizing for shaky videos with various complicated scenes.