Research On Non-rigid 3D Reconstruction Method Based On Multi-objective Optimization

Recovering non-rigid 3D shape and structure from monocular image sequences has many important applications for robot navigation and computer vision. Most of the traditional methods are based on the linear combination model of shape basis to solve the problem of non-rigid 3D motion reconstruction. There are different forms of life in the world, so different objects have different shape basis. This kind of non-specificity of shape basis makes the performance of the traditional non-rigid 3D motion reconstruction algorithm have a large defect, which limits its application and development. Recently, some scholars proposed a non-rigid object 3D motion reconstruction algorithm which based on the combination model of trajectory basis by the duality of temporal and spatial. Due to the trajectory basis can be predefined, so the algorithm in terms of robustness and accuracy have made a great improvement, but it is still far from the practical application requirement.In order to improve the performance of non-rigid object 3D motion reconstruction algorithm and promote the development and application of 3D motion vision technology,this paper has done the following work which based on existing researches.1 、 In the real world, the movement and the shape of an object all have a certain continuity. Based on this consideration, this paper makes a detailed analysis and comparison between real data and the results of non-rigid 3D motion reconstruction algorithm based on trajectory space. The experimental results show that the reconstruction results had a certain gap, which shows that the non-rigid 3D motion reconstruction algorithm based on the combination model of the trajectory has relaxed constraint on the trajectory smoothness and continuity in solving process.2、Aiming at the problem that the non-rigid object 3D motion reconstruction algorithm based on the combination model of trajectory basis, this paper presents a multi-objectiveoptimization algorithm for 3D motion reconstruction. In this method, a multi-objective cost function is constructed, and the function is integrated with the non-rigid motion trajectory smoothness, structural continuity and reprojection error in monocular image sequences.Then, the minimum of the cost function is used as the optimization target for the iterative solution. At the same time, the 3D structure and camera motion parameters also are determined. The experimental results show that the proposed method can improve the accuracy of the reconstruction algorithm and upgrade the motion trajectory smoothness error and structural continuity error and so on.