Research On Electronic Image Stabilization Algorithm Based On Multi-resolution Gray-scale Projection

The imaging process of mine car system is vulnerable to the external environment and carriers’ vibration, so that make the video images sequences unstable. These unstable videos have a great impact on the safe production of mines,subsequent image processing(such as target detection and tracking, etc.), report after accident as well as timely warning and linkage processing,etc.Therefore,the improvement of the stability of images is important for the mines’ safe and efficient production.Electronic image stabilization technology stabilizes video images sequences by the method of image processing. Because of its high accuracy, small size and other advantages, electronic image stabilization technology has been widely used,but there is not yet a general electronic images stabilization algorithm. After analysis and research of main algorithm of motion estimation, motion compensation and motion filtering in electronic image stabilization, this article proposes an electronic image stabilization algorithm suitable for mine car system based on the gray projection algorithm. The main works of this article are as follows:On the phase of motion estimation, due to the gray projection algorithm’s bad performance on images with low contrast and foreground moving objects, a motion estimation method based on multi-resolution gray projection was proposed in this paper. This algorithm firstly pre-processed the images by histogram equalization to reduce the effect of image intensity single contrast, then, solving the problems of local moving objects’ impact on global motion vector by the stratification of multi-resolution, reducing of calculation based on binary search method as well as block most statistical sample method.On the phase of motion compensation and motion filtering, due to the traditional particle filter’s shortcoming of big calculation and long time-consuming, this article proposed a motion filtering algorithm based on estimation window real-time particle filter. By introducing the concept of the estimated window, this algorithm divided the particles set into the subsets with same numbers, then, do particle filter operation on each particle subset. Finally, the final estimation was got by the Weighting of obtained results. For the motion parameters obtained after filtering, their motion compensation was realized through an improved adaptive adjacent frames method.Experimental results show that the algorithms in this article, on the one hand, ensurealgorithms’ accuracy while reducing computation time, on the other hand, solve the problems of low contrast of images and local moving objects’ effect on algorithms’ accuracy, so that it can achieve a long image stabilization.