Laser point cloud scene similarity evaluation method

Abstract

The invention relates to a laser point cloud scene similarity evaluation method, which comprises the following steps of: A1, respectively carrying out semantic segmentation, instance segmentation, node feature coding and node feature aggregation on two frames of laser point cloud scene data Pi and Pj which need to be subjected to similarity evaluation to obtain graph features Ei and Ej of the two frames of laser point cloud scene data; A2, calculating a similarity vector for the graph features Ei and Ej by using a pre-trained neural tensor network V, carrying out dimensionality reduction on the similarity vector through a pre-trained full connection layer to obtain a similarity score Y of the graph features Ei and Ej, and evaluating the similarity of the graph features Ei and Ej according to the similarity score Y. The laser point cloud scene similarity evaluation method has the advantages that the method is more robust and stable for scene changes such as shielding and rotation, the calculation method is simple, the calculation amount is small, the precision is high, and the speed is high.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a laser point cloud scene similarity evaluation method. Background technique [0002] In recent years, simultaneous localization and mapping (SLAM) technology has developed rapidly and plays a key role in robotic applications and autonomous driving. Loop closure detection is an important problem in SLAM, referring to the ability of a robot or moving vehicle to recognize whether it has ever been to a place. This is the most effective way to eliminate the accumulated drift error of the odometry, which can help to build a more accurate map and achieve more accurate positioning. The research of vision-based position recognition has been studied for a long time, and many successful methods have emerged. Most image-based methods work by extracting feature descriptors and then encoding them using methods such as bag-of-words. Related scenes are retrieved by comparing global d...

AI Technical Summary

Problems solved by technology

[0007] At present, there is no laser point cloud scene similarity evaluation method that can solve the above problems at the same time.

Images