Design And Implementation Of Motion And Structure Estimation System Based On Vision-inertia Combination

In recent years,with the improvement of computing power and the rapid development in communication,electrical,mechanical and control fields,the demand for autonomous robots is becoming more and more urgent,especially the robots operating autonomously in complex environment have been the focus of attention.To satisfy the needs of autonomous planning and decision in complex environment,especially in the unknown passive scene,it is of great significance to implement robust,efficient pose estimation and 3D environment modeling.To solve the key problem of autonomous robots,this paper aims to propose a reliable motion and structure estimation system using low cost visual and inertial devices to realize the position and pose estimation and 3D environment modeling under the unknown passive environment.The main contents of this paper can be summarized as follows:First,the consistency analysis of simultaneous localization and mapping problem is given.Based on the observability theory of nonlinear system,the observability of the visual inertial combination system is proved and analyzed.According to the observability analysis of the visual inertial combined system and the consistency analysis of the simultaneous localization and mapping problem,a joint structure of global and local estimator is proposed.Secondly,for low cost inertial and visual components,the mechanism model of inertial and visual components is established,and,the calibration method of the model parameters of inertial and visual components based on maximum likelihood estimation is given,and the calibration method of external parameters based on the non-linear optimization method is proposed.Then,the local feature of the image is modeled based on the sparsity method,and the local feature association and association check method based on the fast matching strategy and the random sampling method are given.The global feature representation of the image based on the word packet model is given,and the image retrieval and the matching formula based on the positive sequence and reverse sequence index structure are proposed.Then,the maximum posterior probability estimation form of motion and structure estimation and the approximate equivalent form of the local state estimation are analyzed and reasoned.The local motion and structure estimation method based on the sliding window optimization is given,and the initialization method of the nonlinear estimation system based on the loosely coupled mechanism is proposed.Global motion and structure estimation method based on pose graph optimization.Finally,the algorithm of the motion and structure estimation system is designed and implemented,and the test and verification of the algorithm based on the standard data set proves the effectiveness of the system method.