A Study Of Global Movement Detection In Stabilization System

With the rapid development of digital image devices, clear and stable video is that people always have been pursued. However, due to the complexity of the applied environment, the jitter is inevitable resulting in unstable and blur, likely to cause visual fatigue and hard to the post-processing. Electronic image stabilization is the technology based on computer and image processing technology to stable shake video, widely used in areas of military reconnaissance system, military fire-control system, civilian aerial system, monitoring system and hand held recording system, it is an indispensable video enhancement technology for moving platform vision system. Reading on the basis of a large number of domestic and foreign literature, proposed the improved algorithms for these complex movements in EIS system.The main contents of this paper are summarized as follows:1. Firstly in this paper we describes the overseas development of EIS system in detail, and analyses the existing problem in stabilization system; Introduce the basic principle, structure of EIS system and mathematical model; then key techniques for motion estimation and motion compensation modules are analyzed such as block matching algorithm, the gray projection algorithm, the phase correlation method, feature tracking method. Then the major algorithms for each module are introduced and compared in real-time processing speed and application scenes; Some experimental simulations and further discussion are made on the existing motion filtering and compensation methods, we also summarize the evaluation criteria for EIS performance.2. In the third chapter, the further discussions about the stabilization issue are made on under the video sequences with foreground interference. Firstly introduces some normal feature detection algorithms, based on the processing speed and applicability, we choose the SURF detection algorithm, for the obtained matching feature points check the validation, we propose a improved the spatial distance criteria is superior to RANSAC the commonly used algorithm, by experiments the validity and the real-time of the improved algorithm are verified; for the foreground interference raises the image stabilization algorithm based on foreground segmentation, proposed a global motion estimation algorithm based on distance criteria and foreground Detection to extract foreground targets, under different video scene experiments test the adaptability and the robustness. The results show that the algorithm can be a good function to suppress the influence of local motion, and output stable video sequences.3.The fourth chapter studies the video image stabilization algorithm with the existence of Depth of Field (DOF) conditions. First, according to the imaging relationship the effects of DOF on camera movement are produced during imaging process, based on the characteristic that local motion vectors on close-range targets is greater than the prospect area, we proposed a algorithm to stabilize the image under the different DOF motion vectors of compensation. At first use of the simple block-matching algorithm to obtain the local motion vectors, based on the fast global K-means algorithm to classify the local motion vectors, and are brought into the Similarity motion model to calculate the different motion vectors of close-range and vista, then compensate the jitter and output stable video sequences. Finally, simulation experiments verify the effectiveness of the algorithm.