Depth Acquire Method Based On Deep Learning And Structured Light

In recent years,the methods of structured light and deep learning are widely used in artificial vision to acquire a depth map of real-world scenes.3D reconstruction technique plays an important role in computer human interaction,telemedicine,3D printing,etc.The research of structured light and deep learning is of great significance to enhance the performance of 3D reconstruction.Two depth acquire schemes which combine structured light and deep learning are proposed in this paper to solve some problems in 3D reconstruction.The depth acquire method which combines phase-shift structured light technique and stereo matching network is proposed in this paper.A deep learning network is designed to work on stereo matching.The coarse depth map which is acquired by stereo matching network is used for phase unwrapping.Then,a fine and accurate depth map is obtained by phase matching process.A depth map with high accuracy and density can be acquired according to this method.Besides,this method can solve the occlusion problem by using both coarse depth information and fine depth information.To evaluate the performance of our proposed method,an experimental platform is established and several experiments are conducted.Quantitative and qualitative experiments demonstrate that the proposed method can generate a high precision depth and relieve the occlusion in the structured light system.The depth acquire method which is based on single-frame composite coding is proposed in this paper.A high hamming distance code table creating scheme is proposed in this method.This kind of code table can enhance the robustness of the performance of shape coding structured light method.Besides,high robustness shape code element is designed in this method to compose the composite coding pattern.In order to enhance the density of depth information,grid line is added to the composite coding pattern.In the process of decoding,we use both morphology method and deep learning method to extract and decode the shape code to improve the accuracy of matching.The recognition network which is used in this method is Densenet.Optimization methods are used in the process of extracting grid lines.Coordinate of the point on grid lines can be determined by the accuracy of pixel in this way.Quantitative and qualitative experiments are provided to show the robustness and accuracy of the proposed depth acquire method.