Monocular vision inertia SLAM (Simultaneous Localization and Mapping) method and system based on adaptive robust kernel

Abstract

The invention discloses a monocular vision inertia SLAM method and system based on an adaptive robust kernel, and belongs to the field of vision robots. The method is mainly improved for monocular initialization, back-end optimization and loopback detection parts, initial feature matching is carried out on a current frame and a reference frame during monocular initialization, two motion models of a homography matrix and a basic matrix are calculated in parallel, the homography matrix or the basic matrix is selected through geometric robust information content, and the motion model of the current frame and the motion model of the basic matrix are calculated in parallel. Carrying out model decomposition and bundle adjustment BA optimization; constructing a cost function for minimizing a re-projection error in a local mapping thread, and applying an adaptive robust kernel to BA optimization; and reducing accumulative errors in a loopback correction process by using a self-adaptive robust kernel in a loopback detection thread. Compared with the prior art, the method has the advantages that the monocular initialization process is improved, the influence of exterior points is reduced by adopting a self-adaptive robust kernel, the optimization effect is enhanced, and the robustness of the system is improved.

Description

technical field [0001] The invention belongs to the field of visual robots, and relates to a monocular visual-inertial SLAM method and system based on an adaptive robust kernel, in particular to monocular initialization, back-end optimization and loop-closure detection part optimization in monocular visual-inertial SLAM. Background technique [0002] With the rapid development of automation integration technology, communication and electronic technology, and high-precision sensor technology, robot technology has become an irreplaceable part of social production and life, and will affect the global automation degree more widely in the future. The robotics industry has been recognized by the world as one of the most important and promising high-tech industries, but it is undeniable that there are still many problems waiting for us to explore whether in the field of robotics or artificial intelligence. Therefore, many experts and scholars have emerged to conduct in-depth resear...

AI Technical Summary

Problems solved by technology

[0005] Nowadays, although the research of visual SLAM has achieved fruitful results in many directions, there are still many problems that need to be optimized and solved. For example, in scenes such as dynamic scenes, missing textures, and complex environments, it is difficult for visual SLAM to run stably for a long time.

The existing monocular vision research mainly has the following deficiencies: (1) In monocular vision, due to the need to normalize the translation vector, resulting in the scale uncertainty of monocular vision, the monocular initialization cannot be obtained from a single frame. Depth information, in addition, when there is only pure rotation in the monocular initialization process, the monocular will not be initialized

(2) In the local mapping thread, although the overall error caused by mismatching and other reasons is reduced, some large errors will still have an unpredictable impact on the result, and it is impossible to improve the accuracy in a breakthrough

(3) In the visual SLAM front-end, whether it is performing pose estimation or mapping, it is done using the relationship between adjacent frames. This algorithm that relies on local constraints will inevitably lead to optimization errors that accumulate frame by frame. resulting in a large error drift

Images