基于双特征和HNSW网络的视觉SLAM闭环检测算法OA

When the current loop closure detection algorithm uses a single feature to characterize the image,the effective information of the image is insufficiently characterized,and there is an excessively long time-consuming in the image loop closure search,resulting in low loop closure detection accuracy and poor realtime performance.Therefore,this paper proposes a loop closure detection algorithm based on ResNet18 convolution features and traditional ORB artificial features combined with hierarchical navigable small world networks.Firstly,the ResNet18 convolution features and ORB artificial features were extracted from the detected image.The extracted features were encoded by the constructed visual dictionary using the VLAD algorithm,the two VLAD codes were fused by the optimal fusion weight co-efficient,Secondly,the VLAD index structure constructed by the hierarchical navigable small world HNSW network was used to accelerate the retrieval of the fused VLAD coding.The results show that:Compared with the VLAD coding based on single ResNet18 convolution features,the algorithm based on dual features fusion VLAD coding improves the accuracy of loop closure detection on NewCollege and CityCentre datasets by 29％and 20％respectively.Compared with the global traversal search of the original algorithm,the total loop closure search time on the NewCollege dataset using the VLAD in-dex structure search algorithm of the HNSW network is reduced by 51％,and the accuracy of loop clo-sure detection was as high as 85％.The loop closure detection algorithm based on dual features and HNSW network has high loop closure detection accuracy and short search time,which provides a new design scheme for high precision realtime loop closure detection system.