Visual-inertial SLAM algorithm based on enhanced feature extraction and sensor coupling

• Main Author: Zheng, Yumin
• Other Authors: Xie Lihua
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Electrical and electronic engineering::Control and instrumentation::Robotics
• Online Access: https://hdl.handle.net/10356/158541
• Institution: Nanyang Technological University

Service robots can facilitate people’s lives and perform some highly repetitive tasks. One of the core algorithms of such robots is Simultaneous Localization and Mapping (SLAM) based on vision. However, due to the characteristics of the camera itself, it cannot effectively extract image features in weak texture and violent motion scenes, resulting in loss of tracking. The Inertial Measurement Unit (IMU) has a better estimation value when moving rapidly, which can make up for the lack of visual sensors. Therefore, this dissertation studies SLAM technology based on the method of fusion of vision and inertial, and proposes a more effective method for extracting features uniformly and creating word labels, which can accelerate the speed of frame matching. The fusion of IMU and Camera combines both loose-coupled and tight-coupled methods. The proposed method takes the pose estimation of pure visual tracking and IMU pre-integration results as the prior conditions to get more accurate initialization measurement. After the initialization is completed, the graph optimization method is used for tight-coupled optimization, which improves the tracking accuracy of the system. The proposed method is tested under open source datasets and achieves better performance than some of the open source SLAM algorithms.