Loop Closure Detection for Visual SLAM Systems Using Various CNN algorithms Contrasts

In the Visual Simultaneous Localization and Mapping (V-SLAM) system, loop closure detection is playing a decisive role in the accurate construction of maps. Traditional loop closure detection mainly use the Bag-of-Words (BoW) model to extract features and judge the similarity of images. Recently, more related researches tend to combine machine learning with loop closure detection, but so far no breakthrough results have been achieved yet, and even it is still in the stage of finding suitable machine learning algorithms. In this paper, by improving the loop closure detection process, a variety of CNN modified algorithms (including VGG, Inception 1-4, ResNet, Inception Resnet, etc.) are used in the loop closure detection, and are further conducted for comparison and analysis. Through a number of experiments and data analysis, the advantages and disadvantages of these machine learning algorithms in V-SLAM loop closure detection are finally obtained, which would provide a solid foundation for the research work in the future.