Research And Application Of 3D Object Recognition For Indoor Scenes

With the development of computer storage technology and depth sensor technology,the sensor can obtain high-quality color information and depth information at the same time.By using the complementary advantages of color and depth information,it is expected to solve many problems of 2D image recognition.Therefore,this paper focuses on the key technologies(such as depth map repair,3D feature extraction,and classification algorithm)of 3D object recognition to improve the accuracy of object recognition.The main research contents are as follow: Considering the depth map captured by Kinect suffer from the problems of hole and noise,this paper proposed a depth recovery method utilizing valid support region.For a pixel in depth map hole,calculate its depth value using its known depth values of valid support region that is obtained by cavity type determination and local area segmentation.In the feature level,combining the advantages above color information and depth information,the paper extracted the SIFT feature from color map and 3D shape feature from the depth map as feature descriptors of indoor scene objects.On the decidion level,in view of the small sample classification capability of the support vector machine and the feature automatic learning characteristics of the convolutional neural network,the paper established two model of 3D object recognition for indoor scene based on support vector machine and convolutional neural network.Experimental results show that the proposed recognition algorithm improves the accuracy of 3D object recognition for indoor scene.