Research on Multi-depth-of-field Electronic Image Stabilization Algorithm for In-Vehicle Videos

Abstract

The vehicle-mounted cameras provide video streaming information for the post-processing phase of the intelligent vehicle, such as control and decision making. According to the characteristics of in-vehicle videos, this paper studies a video stabilization algorithm considering different depths of field. On the motion estimation stage, for the problem of inaccurate estimation inter-frame motion vectors caused by different depths of field of feature points, this paper proposes a method to repair inter-frame motion vectors. The method can correct the feature point position of the current frame and improve the estimation accuracy of the global motion vector. In the motion smoothing phase, a real-time online optimization framework and adaptive weights are used to balance the relationship between motion smoothing and motion following. Experiments prove that the accuracy of motion estimation of the algorithm in this paper is better than other improved algorithm, and the average value of PSNR after video stabilization is increased by more than 3dB compared with the original videos.