Keyframe-Based Visual-Inertial SLAM Algorithm

The problem of SLAM(Simultaneous Localization and Mapping)is a basic problem in research field of mobile robotics as well as a key step to implement the independent navigation and autonomous control system.This thesis presents a keyframe-based visual-inertial simultaneous localization and mapping algorithm for monocular and stereo cameras.The main research contents include:Focus on the problem that the accuracy and robustness of the current visual-inertial simultaneous localization and mapping algorithm are not high due to lack of the loop-closing detection,the visual-inertial SLAM system provides locally consistent trajectory through a visual-inertial odometry.The global consistent trajectory is achieved through finding the matching keyframes with image retrieval algorithm to detect the closed loop and creating a new nonlinear optimization equation to optimize the detected closed loop in parallel,which can detect the global closed loop and global optimization.The problem of the current algorithm which the map is built with single message is avoided with the continuous global map is created.This algorithm realizes the goal of building a global map through the dual constraints of visual inertial information and closed loop.Focus on the problem that the current visual location algorithm can only achieve the relocation of equipment,without having the ability to continue the subsequent building,our algorithm use the image retrieval algorithm as introduced in closed-loop detection to find a series of consecutive matching key frames as the relocation hypothesis,thus find the relationship between the new keyframes in the map and the constructed map's keyframes.Then continue the subsequent building according to the the relationship between the new keyframes and the constructed map's keyframes.The experiments are carried out using the EuRoC dataset to evaluate the accuracy,relocation capabilities and run time of the algorithm.Experiment results show that,compared with the current visual-inertial SLAM algorithm,the proposed method can reduce error accumulation and the drift,relocate the camera and continue to build the map on the basis of the previous map under the conditions of real-time.