Object recognition methods based on RGB-D images

Abstract

Object recognition is one of the main contents of machine vision research. RGB-D image, which integrates depth features, provides a new way for accuracy improving of object recognition. In order to obtain its complete summary of many achievements in recent years. Firstly, the research of object recognition based on RGB-D image was summarized, and the advantages of RGB-D object recognition combined with depth image were discussed. Secondly, based on the analysis of the sample characteristics of RGB-D dataset, the characteristics of the early manual feature recognition method based on early knowledge and the RGB-D object recognition method based on feature learning were summarized and analyzed, and the multi-modal fusion method of RGB-D data was discussed in detail. Finally, based on the Washington RGB-D object dataset, the recognition results of more than 20 models and encoding methods were systematically compared and analyzed, and the trend and future research direction of RGB-D object recognition were prospected.

Keywords

• rgb-d image
• object recognition
• 3d feature
• depth image
• multi-modal