Panoramic inertial navigation SLAM method based on multiple key frames

Abstract

The invention discloses a panoramic inertial navigation SLAM method based on multiple key frames. The method comprises the following steps of: building a multi-target panoramic visual model, combiningan inertial navigation unit and tightly coupling a fusion visual feature, and realizing a high-precision positioning and a scene map through multiple key frames nonlinear optimization. The method isused for solving the problems in the prior art of the insufficient information acquisition by environment-aware methods and the uncertainty of fast rotation, and limited view limitation, illuminationinfluence and visual degradation of an existing visual SLAM method, providing an effective automation method for a robot environment modeling and positioning navigation in a complex environment such as strong light / weak texture / motion blur, improving the accuracy and the robustness of the method, so that ensuring the sufficiency of autonomous navigation of the robot, and considering the calculation efficiency, can achieve the real-time performance of the visual SLAM method.

Description

technical field [0001] The invention belongs to the field of computer vision detection, relates to a visual perception SLAM method, in particular to a multi-keyframe-based panoramic inertial navigation SLAM method. Background technique [0002] SLAM is the key to realizing the autonomy of mobile platforms. In recent years, it has shown great application value in many fields, especially navigation robots and virtual reality. Simultaneous Localization and Mapping (SLAM) refers to a method in which the mobile platform uses the body or external sensors to perceive and construct a consistent environmental map and determine its own relative pose when the pose is unknown. Compared with laser SLAM, visual SLAM can obtain unstructured spatial information, provide semantic annotation scenes, and have greater application significance in complex environments. For this reason, to achieve the robustness and accuracy of visual SLAM, a general and real-time visual SLAM method is needed. ...

AI Technical Summary

Problems solved by technology

[0004] At present, traditional vision methods are based on RGBD cameras to obtain depth data and then convert them into two-dimensional laser data, which does not make full use of visual information, nor is it a true visual SLAM method framework, such as the patent application number 201710835285.4 "A Kinect-based Indoor mobile robot visual SLAM method"; most scholars are also studying monocular SLAM, but this method only uses a single camera, has a limited field of view, and cannot directly estimate the distance of the scene features, such as the patent application number 201710658521.X " Monocular Visual SLAM Algorithm Based on Semi-Direct Method and Sliding Window Optimization"; the document "ORB-SLAM: a versatile and accurate monocular SLAM system" proposes a multi-threaded visual SLAM method based on key frames, using ORB features with high stability and Taking into account real-time, but sensitive to light, easy to lose in extreme situations such as fast rotation

None of the above patents and documents solve the problem of limited field of view, and cannot be applied to weak texture environments. At the same time, when the vision is degraded (motion blur, large scenes), the scale estimation has a greater impact, resulting in the failure of the normal use of the visual SLAM method.

[0005] Therefore, how to provide a stable and reliable visual SLAM method to solve the insufficient information obtained by the environmental perception method in the prior art and the uncertainty of rapid rotation, and to ensure the accuracy and stability of the visual SLAM method has become an art field. technical issues to be resolved

Images