Single-shot Grid Coding Based Structured Light Method

Along with the technology develops,people are no longer satisfied with cameras which can only acquire two-dimensional images.Depth acquisition has gradually appeared in our daily life,such as human-computer interaction,virtual reality,and medical instruments etc..As one of the depth sensing and reconstruction approaches,the structured light methods are widely used.In this thesis,two structured light methods which combine line grid and M array are proposed.Firstly,we introduce the principle and physical model of structured light system.We also explain the coordinate transformation in structured light.To tackle the problem of tradition grid structured light pattern,a hybrid encoding method that combines line grid and M array is proposed.The M array is embedded in the grid.The location of graph code is determined by line extraction.The decoded graph can be used to solve the correspondences between the projected pattern and the acquired image.The proposed method can achieve high accurate depth map without increasing the number of pattern.Practical experiments validate this method.In order to attain denser depth image,a modified grid pattern is proposed.In this modified method,we adopt the epipolar constraints to convert the two-dimensional M array into onedimensional sequence,which results in the decrease of the size of the window and the number of code words.Because the Hamming distance of graph code is improved,the proposed method is robust.Comparative experiments between the two proposed structured light methods prove that the second approach can achieve more reliable and denser depth image.In order to realize real-time depth sensing of dynamic scenes,we use GPU to accelerate the two proposed methods.We describe the details of the CUDA design process.Results of the GPU processing time is given in Chapter 5.The results show that the GPU parallel computing can effectively speed up the calculation of the proposed methods.Experimental results demonstrate that the proposed methods which can acquire high accurate and realtime depth measurement are suitable for indoor scenes.At the end of this thesis,we summarize all our words and prospect the further work.