基于深度学习的语义特征点提取匹配算法

Deep Learning-Based Semantic Feature Point Extraction and Matching Algorithm

摘要: 针对传统ORB算法在特征点提取匹配方面存在的误提取和误匹配问题，以及在语义SLAM应用方面的不足，本文提出了一种基于深度学习的语义特征点提取匹配算法。首先，利用语义分割网络获取匹配图像的语义信息，再结合四叉树分配及自适应提取阈值算法，以实现提取均匀分布的语义特征点。其次，为了剔除特征点匹配环节中存在的误匹配，本文设计了一种语义角度直方图匹配优化算法，通过结合匹配语义特征点的角度和语义信息用于剔除误匹配，从而获得准确的匹配特征点对。最后，通过对比实验评估本文所提算法，实验结果表明论文提出的语义特征点算法能够有效消除误匹配并实现提取特征点的均匀分布，证明了本文算法在多种场景下的优越性能，满足语义SLAM系统的需求。

Abstract: To address the issues of mis-extraction and mis-matching in traditional ORB algorithms for feature point extraction and matching, as well as the inadequacies in semantic SLAM applications, this paper proposes a semantic feature point extraction and matching algorithm based on deep learning. Firstly, the semantic information of matching images is obtained using a semantic segmentation network, and then combined with a quad-tree allocation and adaptive extraction threshold algorithm to achieve uniformly distributed semantic feature points. Secondly, to eliminate mis-matches in the feature point matching process, a semantic angle histogram matching optimization algorithm is designed, which combines the angle and semantic information of matching semantic feature points to eliminate mis-matches, thus obtaining accurate matching feature point pairs. Finally, through comparative experiments, the results demonstrate that the proposed semantic feature point algorithm effectively eliminates mis-matches and achieves a uniform distribution of extracted feature points, proving the superior performance of the algorithm in various scenarios and meeting the requirements of semantic SLAM systems.