Global Motion Estimation And Motion Filtering In Panoramic Electronic Image Stabilization

Panoramic electronic image stabilization in the field of electronic photography has been widely used,but also can be extended to automotive electronics,intelligent transportation and other video surveillance.Multi-CCD panoramic image stabilization can effectively improve the compression rate of video,and suppress the inter-frame wave jitter due to the instability of the image carrier,and alleviate the visual fatigue of the observer.In this paper,global motion estimation and motion filtering in panoramic electronic image stabilization are studied.Each camera is made a spherical projection respectively.Key camera is set,and each camera's sphere will rotate to the sphere where the key camera is located.Through motion estimation of the key camera,the whole spherical trajectory will be obtained.In the process of global motion estimation.Firstly,we find the camera with the least moving objects in the scene as the key camera according to the visual dense optical flow of each camera.And then the image of the key camera is divided into foreground and background so that the feature points are selected on the background and the global motion estimation accuracy is improved.The motion estimation is carried out using the ICP algorithm,and the obtained motion trajectory of the panoramic camera system is represented by rotation matrix.Secondly,the rotation matrix sequence is filtered by improved IIR filtering based on manifold.Thirdly,the smoothing motion trajectory is compared with the original motion trajectory to obtain the rotation matrix which causes the panoramic camera system to shake.Fourthly,rotate the sphere in the opposite direction along the rotation matrix,and finally expand the spherical surface to obtain a stable panoramic image sequence.The experimental results show that the global motion estimation method of multi-CCD panoramic camera can effectively suppress the local motion interference,and select the camera with the least moving objects in the scene as the key camera,and can better distinguish between the foreground and background,so that feature points are selected on the background.The process of filtering the motion trajectory of the panoramic camera system can effectively reduce the random jitter and preserve the active motion to achieve the effect of path smoothing without deviating from the original trajectory.The proposed algorithm can effectively solve the problem of motion estimation and motion filtering in image stabilization of panoramic camera.