Self-positioning error elimination method for closed-loop detection robot based on effective key frame

Abstract

The invention discloses a self-positioning error elimination method for a closed loop detection robot based on effective key frames. The method comprises the steps of coarsely extracting the key frames based on a drift rate, wherein the drift rate is used as the other selection principle to be fused into key frame extraction of ORB-SLAM; adopting fine extraction based on deep learning for the keyframe parts of which regional characteristics are similar, using Alex-NET to learn, and thus completing fine extraction of the key frames; and at last, performing historical closed loop detection based on the set of the aforementioned effective key frames, to judge that whether a robot enters a same historical state. According to method, the motion situation with large drift can be well processedvia a multi-angle effective key frame selection strategy; by using a deep learning network the problem that key frame extraction of the similar areas is fuzzy is avoided; and due to closed loop detection based on the historical key frame data set, the problem of time waste caused under the condition that the closed loop occurs or the closed loop less occurs is avoided, the running speed of the system is improved, and the algorithm complexity of the whole process is reduced.

Description

technical field [0001] The invention relates to a robot closed-loop detection self-positioning error elimination method based on an effective key frame, in particular to a closed-loop detection method under dynamic offset conditions. Background technique [0002] For a long time, people have used robots to measure and construct complex environments. Figure 1 It has not been ideal, and the following problems of feature quantities under dynamic conditions are poor, the environment texture is sparse, the calculation load is large, and the real-time performance is poor, which all affect the measurement effect. Therefore, a series of robot SLAM composition schemes have been proposed. [0003] The robot SLAM algorithm initially uses a filtering method, such as a method based on Kalman filtering, but it must be based on the Gaussian assumption, which greatly limits the application occasions. In 2007, Montemerlo used the particle filter to implement the FastSLAM algorithm. Although...

AI Technical Summary

Problems solved by technology

[0005] Purpose of the invention: In order to solve the problem of loss of effective key frames of SLAM in complex dynamic environments and the problem of excessive algorithm complexity caused by frame-by-frame closed-loop detection in the prior art, the present invention provides an automatic closed-loop detection of robots based on effective key frames. Positioning Error Elimination Method

Images