Semantic map construction method based on convolutional neural network and computer storage medium

Abstract

The invention provides a semantic map construction method based on a convolutional neural network and a computer storage medium, and the method comprises the following steps: S1, receiving a 2D image,transmitting the 2D image to a convolutional neural network model, and outputting neurons of dense pixel-level semantic probability map points; S2, tracking classification probability distribution ofeach curved surface by adopting a Bayesian updating model; S3, providing data by adopting an ElasticFusion method to carry out relevance prediction, and updating probability distribution; and S4, improving semantic detection through the scale information of the map by utilizing a conditional random field regularization model. According to the semantic map construction method based on the convolutional neural network, semantic segmentation can be carried out based on the convolutional neural network, the semantic map is generated, and the robustness of the semantic map under few weak texturesis enhanced.

Description

technical field [0001] The present invention relates to the field of navigation systems, more specifically, to a navigation system map construction method, and more specifically, to a convolutional neural network-based semantic map construction method and a computer storage medium. Background technique [0002] The prior art discloses a point cloud semantic map construction method based on deep learning and lidar. This technical solution describes a deep learning based on lidar features and a construction method of semantic map, that is, scanning point clouds based on lidar , using K-nearest neighbors or similar methods for unsupervised learning to expand data, and then introduce convolutional neural networks for semantic recognition. After recognition, semantically label the geometric map formed by lidar, and then construct a corresponding map. In addition, a series of algorithms such as orb-slam and svo have proposed methods based on bag-of-words (BoW) or semi-dense featur...

AI Technical Summary

Problems solved by technology

Simple geometric features or low-dimensional naked-eye features contain too little information, and cannot directly extract enough information to build a more accurate semantic map

Images