A Visual SLAM Algorithm Based on Multi-Feature Optimization for Indoor Environments

How to locate mobile robots, unmanned aerial vehicles and other carriers in indoor environment with GPS signal rejection is one of the key issues to determine the completion of their tasks. This paper uses the continuous image information obtained by the camera to design and implement a visual SLAM algorithm based on multi-feature optimization for indoor positioning scenes. In order to reduce the tracking failure caused by insufficient features, this paper adds line features, uses LSD segment extraction algorithm to extract and process line features from video frame images, uses line features to combine edge point features and corner points to initialize the algorithm, and then tracks in adjacent frames and local maps. In addition, a priori model of uniform motion is used, which takes into account both the dynamic and the stability. Finally, for indoor scenes, this paper will test the accuracy, robustness and real-time performance of the algorithm compared with existing algorithms in different datasets and real-time environments through indoor dataset experiments and camera experiments in real-time indoor scenes, and verify that the algorithm can be applied in real-time environments, and has good robustness, accuracy and real-time performance.