Research On The Key Technologies Of RGB-D Image Processing Based On Deep Learning

With the emergence of a new generation of RGB-D cameras,people can obtain color and depth images more conveniently and process the stereo images effectively.At this stage,the technology of stereo image processing has developed rapidly and achieved remarkable results.Especially,the advent of consumer depth cameras has brought new vitality to the field of stereo image processing.Under this background,the technologies of depth map super resolution and RGB-D object recognition are deeply studied in this paper.A color-guided depth map super resolution method using convolutional neural network is proposed in this paper.First,a dual-stream convolutional neural network,which integrates the color and depth information simultaneously,is proposed for depth map super resolution.Then,the optimized edge map generated by the high resolution color image and low resolution depth map is used as additional information to refine the object boundary in the depth map.Experimental results show that the proposed method improves the quality of depth map super resolution.A RGB-D object recognition method based on convolutional-recursive neural network and fisher vector is also proposed in this paper.The method uses color and depth information,and adopts multi data models feature extraction method.For the original color image and depth map,the convolutional-recursive neural network and convolutional-fisher vector-recursive neural network are used to exact the texture and shape features respectively.In order to capture more comprehensive features for object recognition,the gray image and the surface normal are introduced in our model,and the convolutional-recursive neural network is utilized to explore the corresponding features.At last,these four features extracted from different data modalities are integrated into the softmax classifier to achieve RGB-D object recognition.The proposed algorithm is verified in the standard RGB-D database.Experimental results show that the proposed algorithm can effectively improve the object recognition rate.