Study On Algorithms Of Electrical Image Stabilization Based On Monocular Vision

In recent years, the technology of computer vision has been growing fast as this advanced technology has been applied in the field of automated driving system and intelligent monitoring. Among them, digital image stabilization is becoming a hot topic and attracting more and more attentions because of its low cost and high accuracy.In our research, we studied the digital image stabilization technology based on monocular vision system. Aiming at obtaining accuracy motion vectors, we studied two different kinds of methods which has great improvement than previous algorithm in terms of real time performance and accuracy. The mainly studied results of this dissertation are shown as follows:Firstly, we simulated compared RPM, BMA and PA algorithms and analyzed the advantages and disadvantages of different algorithms, thus providing theoretical basis for the following research.Secondly, aiming at improving the real time performance and robustness of traditional block matching algorithm, we come up with adaptive search strategy, SSDA, sparsity strategy and voting strategy to modify the disadvantages of low accuracy and high computation.Finally, we studied different optical flow algorithms and compared the results of Moravec corner point detection algorithm, Harris corner point detection algorithm and Fast corner point detection algorithm on the basis of simulation. We then picked out proper corner point detection algorithm and combine it with optical flow theory to make image stabilization. In this process, we applied RANSAC to eliminate invalid corner points to enhance the robustness of our algorithm; we applied Gaussian Pyramid to make Multi-Scale Space Transformation for each frame to improve the real time performance; we applied Affine Transformation in the estimation of motion vectors to solve the problem of rotation.The algorithm we researched is developed in the environment of VS 2010 and OpenCV 2.4.9 using C++. Some part of the research is simulated in MATLAB. The experimental results show that our algorithm achieved a high performance in terms of real time performance and stability and robustness, it has a great prospect in the field of automated driving system and intelligent monitoring etc.