A visual SLAM positioning method based on semantic prior and probability propagation

By using semantic segmentation and Bayesian probability models in the SLAM system, a prior probability of feature points is given and dynamic probability is updated, and the problem of low positioning accuracy in a dynamic environment is solved, and higher positioning accuracy and robustness are achieved.

CN114022768BActive Publication Date: 2025-03-28SOUTHEAST UNIV
2 Cites 0 Cited by

Patent Information

Application Number
CN202111302614.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-11-04
Publication Date
2025-03-28
Estimated Expiration
2041-11-04

AI Technical Summary

Technical Problem

Traditional SLAM technology has reduced positioning accuracy in dynamic environments, and existing methods fail to fully utilize the static situation of feature points when dynamic objects move and stop.

Method used

The visual SLAM positioning method based on semantic priors and probability propagation is adopted, and semantic segmentation is performed through the MASK R-CNN neural network, which gives different prior probabilities of feature points under different semantic masks. The Bayesian probability model is used to update the dynamic probability of feature points, eliminate dynamic feature points and pass dynamic probability.

Benefits of technology

It improves the positioning accuracy and robustness of the SLAM system in a dynamic environment, can detect static feature points of dynamic objects, and enhances the accuracy of position estimation.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The present invention discloses a visual SLAM positioning method based on semantic prior and probability propagation, including: firstly, using MASK R-CNN neural network to semantically segment input images, obtain semantic masks of various objects, and then classify different semantic masks, and assign different prior probabilities to feature points under different semantic masks and non-semantic masks; secondly, extracting feature points from the input image, using improved epipolar geometric constraint method to perform mobile consistency detection, and obtain dynamic probability under geometric model; then using Bayesian probability model, updating the probability of feature points under semantic mask and non-semantic mask at intervals of specified frames, and selecting key frames; finally, only performing BA optimization on feature points whose probability on key frames is less than a certain threshold, and obtaining camera posture. The present invention can detect some static feature points existing between frames of multiple dynamic objects, and can effectively improve the accuracy and robustness of positioning of visual SLAM system in dynamic environment.
Need to check novelty before this filing date? Find Prior Art