Camera Parameter Optimization Based On Motion Smoothness Constraint

Computer Vision is a simulation of human vision using computer and relevant facilities, with the target to make computers and other intelligent machines artificially intellegent and efficient in vision. In the domain of computer vision, as an important concept, camera parameter depicts the mutual relations between the geometrical object location and its corresponding projection in the image. Camera parameter plays an important part in the domain of3-D reconstruction and augmented reality. The accuracy and the continuity of camera parameter affects their performances. Because of the impact of CCD quantization error and feature detection error, the computed camera parameter of every image is contaminated by errors. For an image sequence with the camera moving smoothly, there are abrupt changes and fluctuations of the camera parameter that should change continuously. Traditionally, optimization approaches for camera parameter focused on minimizing projection error without considering the continuity of parameter. Therefore the camera parameter after optimization still has breaks and acute fluctuations, which badly affect the performance of augmented reality applications.Smoothness is an inherent characteristic of camera motion trajectory in the professional video shooting. We could apply this constraint to the optimization of camera parameter in order to improve the continuity of parameters.In allusion to the continuity problem of camera parameter sequence, this paper presents a novel approach of parameter optimization based on the smoothness constraint, using the smoothness of camera motion as a constraint. First we analyse the characteristic of motion smoothness and describe it mathematically with the method of curve fitting. Then we give the new optimization method with the mathematical constraint of smoothness for postprocess and real-time process of video sequences. Finally we choose some videos and process them with the new optimization method in the augmented reality system for Shaanxi TV Station. The experiment result demonstrates its performance. The method improves observably the continuity of camera parameter sequence while maintaining the parameter accuracy.