Abstract

In this paper, we first study the implementation process of the improved VSLAM algorithm. After initializing the root node containing all the features, a quadtree node partitioning is performed on the image and the network grid of each layer of the ginormous tower. And FAST feature points are extracted by adaptive thresholding within the grid and a specific number of feature points are saved according to the size of Harris response values within the grid. Then the algorithm is experimentally tested for feature point distribution uniformity, matching performance and SLAM performance, and compared with the false matching coarse screening algorithm. Finally, the improved VSLAM algorithm is applied to extract and analyze the symbolic representations, image rhetoric and visual segment rhetoric of Luoshan shadow. In terms of representation matching, the VSLAM algorithm matches at 50%-60%, and the improved matching algorithm matches at 70%-80% with an accuracy rate of 20%. In terms of image rhetoric, 87 shadow works use anthropomorphic rhetoric, the number of works using repetition, pun, reference and exaggeration rhetoric is about 100, and the largest percentage of examples is about 220, and 100 shadow works use accumulation and comparison rhetoric. The improved VSLAM algorithm can be good for the extraction of Luoshan shadow representations and rhetoric, which is beneficial to the study of shadow.