Research On Electronic Image Stabilization Technology Based On Adaptive Hybrid Filtering

With the rapid development of modern society, the camera system is becoming more and more popular, but they are often in a shaking state when taking photos, so the video have jitter.The jitter video can cause people to produce visual fatigue, affect the subsequent operation of the video. The principle of electronic image stabilization technology is using the technology of digital signal processing, because its advantages is high precision, low power consumption and low cost, the domestic and foreign experts are more interested in electronic image stabilization.The accuracy of the existing image stabilization technology is not high, we have to study how to improve the stabilization precision. Among them, the motion filter is an important part of the stabilization, the kalman filter is updated in real time, deal with the noise fastly, many times of iteration, good robustness, strong stability and other advantages, the electronic image stabilization has been widely studied. In this paper, combing with the actual scene, the kalman filter is studied and improved.the main contents are as follows:(1) A detailed study of the various methods of global motion estimation, focusing on the global motion estimation based on SIFT feature matching of SIFT feature matching, and related experiments are conducted, the research is how to combine it with the kalman filter algorithm, the analysis of global motion estimation vector process;(2) Focus on the study of kalman filter algorithm and its improved, according to the kalman filter is incomplete removal of noise. because of low pass filter can be obtained according to the cut-off frequency of filter from spectral analysis of the original data, and put forward a kind of hybrid filtering method based on kalman filter and low pass filter with a Hanning window are combined, the algorithm is globally the motion vectors were estimated by the kalman filter, then estimate the motion vector is input to the low pass filter, to smooth image. The experimental results show that the hybrid filter for video stabilization effect than kalman filter is good, more in line with the requirements of the people.(3) The cut-off frequency of hybrid filter algorithm is not according to each frame of video motion estimation to adjust the low pass filter, resulting in inaccurate estimation precision, and by the adaptive filter theory, method of adaptive mixed filter based on hybrid filter on the cut-off frequency, automatically change the low pass filter the algorithm mainly based on video jitter. The experimental results show that the adaptive hybrid filter is better than other filtering methods including translation and rotation jitter video complex.