Research On Object Recognition Method Based On RGB-D Features Fusion

Object recognition is the basis of people's understanding of image content and is widely used in important fields such as image retrieval,video security and artificial intelligence.In recent years,the research of object recognition technology has made great progress,but there are still some shortcomings in the description of object distance and how to establish a simple and effective object recognition method.Therefore,the in-depth study of object recognition has important practical significance and application prospect.In recent years,the rapid development of depth information acquisition equipment has created conditions for the majority of researchers to solve the problem of object recognition.In this paper,object recognition is studied based on Intel RealSense D435 depth camera.The work of this paper is mainly divided into the following aspects:Firstly,the object recognition system based on depth feature and color feature is constructed,and the imaging principle of D435 depth camera is analyzed.The depth data information and RGB data stream are correlated and processed,and the depth image and RGB image of position matching are formed.Secondly,color feature and depth feature are extracted.First,get the color feature of the object through RGB image.By analyzing the characteristics of the color space,HSV color space conforming to human vision system is selected,and the color feature description of the object is carried out.Then,the object is characterized with color features,and the object's color feature codes are stored in the sample database.Second,the depth features of the object are acquired through the depth image.Aiming at the noise problem of the depth image acquired by depth camera D435,the main noise types of the image are analyzed.The nearest neighbor interpolation method is used to repair the empty area in the depth image and the median filtering algorithm is used to filter the edge burr of the image.Then,according to the depth difference between the target object and the background,the target object area is obtained,and the object shape feature is represented by the ratio of object pixel to background pixel.In the meanwhile,the actual size feature of the object is obtained by the relationship between object depth and object imaging.The last,the object's shape and size are used to describe the object's depth feature,and the depth features are used to encode the object's features,and the depth features code is stored in the sample database.Finally,an object recognition system is established by combining color feature and depth feature.The object to be measured is matched with the object in the sample database and the target object is output.The results of experiment show that the recognition rate of object based on fused feature is improved by nearly 10%,which is obviously higher than that of object based on single feature.Compared with the existing deep learning recognition methods and core descriptor recognition methods,the recognition method in this paper is simpler.The feature extraction and description process is simpler and the object recognition rate is higher.