| With the advent of the Io T(Internet of Things)era,large-scale indoor scenes will continue to increase,and people's perception of the environment in indoor scenes will be also increase.In order to solve the problem of fast retrieval and accurate matching of Visual Map when users locate online,indoor scene classification has attracted more and more attention.As an extension of scene classification,indoor scene classification can not only improve the efficiency of image retrieval,but also has a wide range of applications in areas such as intelligent security and robot navigation.Aiming at the complexity of the indoor scene structure and the high similarity between scene class features,an effective indoor scene classification scheme based on multiple image descriptor is proposed.This scheme effectively compensates for the lack of dimensional information when color images capture 3D object information as 2D visual information during the perception process by introducing depth images.Our scheme mainly includes three parts: spatial segmentation model,greedy descriptor filtering algorithm,and descriptor fusion mechanism.Through the joint application of these three parts,the information of the two descriptors is effectively integrated into each other to achieve the generation and optimization of indoor image descriptors.Performance analysis and simulation results show that indoor scene classification scheme based on multiple image descriptor can effectively improve the quality and efficiency of the descriptor,and its classification effect of fusion descriptor is always better than single descriptor.Especially in the face of medium or large descriptor,the classification effect has far exceeded the traditional Principal Component Analysis(PCA)algorithm.At the same time,compared with other classification algorithms that fuse depth information,our scheme effectively improves the indoor scene classification accuracy without increasing the running time. |
PDF Full Text Request