Research And Realization Of Electronic Image Stabilization Technique For Jitter Video

As video cameras are widely used in people’s life, video stabilization begins to attract people’s attention. When we are shooting with video camera, the unstabilization of the camera may lead to a vibration of the objects in the video, which will makes the observers tired and result in the decrease of judging accuracy, so we need to stabilize the video. Thanks to the development of computer image processing and large-scale integrated circuits, research of video stabilization has turned into the field of electronical video stabilization. The purpose of electronical video stabilization is to remove the instability between frames and reconstruct a stabilized video, by detecting the inter-frame motion vector and image compensation.To improve the efficiency of the video stabilization system, an algorithm based on Harris and modified Hu-moments is proposed in this paper, whose goal is to reduce the computational complexity while keeping the processing effect of rotation cases, focusing on the modules of motion estimation and smoothing, considering the characteristic that the shifting and rotation is fierce while the scaling is slight in unstable videos. The main work of this thesis is as follows:Firstly, an algorithm based on Harris and modified Hu-moments is proposed in this paper. In the algorithm, Harris corners of each frame are detected as the feature-points, and the modified Hu-moments in the corner-centered neighborhood are calculated to be the feature-vectors corresponding to the feature-points. The affine model is used in the motion estimation between the frames. After getting the motion traces of each direction, the Gaussian filter is adopted to remove the unwanted camera movements. Finally, a stabilized video sequence is generated after compensation being exerted on the frames.Secondly, the performance evaluation of the algorithm proposed in the thesis is taken by simulation experiment. We choose an algorithm based on Harris corner block feature and an algorithm SIFT(Scale Invariant Feature Transform) feature as the comparatives, and evaluate the performance in aspects such as feature-matching accuracy, computational complexity and time-consuming of feature-detecting, feature-extracting and feature-matching. Thirdly, on consideration of practical applications, we write the core code of the video stabilization application. Based on the OpenCV library, coding of the core modules such as motion estimation, smoothing and compensation is accomplished with C/C++, which is used in the Linux application program, and the framework and interface design for Android application are also completed, together with the transplantation of the core module, which can support the future development for hand-hold Android devices on framework and core library.The experimental results indicate that when the images are rotated or slightly zoomed, the compute complexity is greatly reduced in the processing of feature-matching with high quality results. In addition, the C/C++code and the Linux application can also work well functionally with ideal effects.