Research On Depth Information Acquisition Technology Based On ToF-Binocular Fusion

The acquisition of depth information is one of the key technologies for 3D applications.In recent years,with the continuous expansion of 3D technology,the fields of human-computer interaction,automatic driving,3D reconstruction and intelligent monitoring have higher requirements for the accuracy of the depth acquisition technology and the adaptability of the scene.Therefore,improving the accuracy and robustness of the depth acquisition technology has become an urgent problem in the field of 3D sensing.Currently,among the many methods of deep information acquisition technology,binocular stereo vision can be regarded as representative of passive deep acquisition technology,and ToF can be regarded as representative of active deep acquisition technology.Inevitably there are some defects in using these two technologies alone.At the same time,the two also have the characteristics of complementary advantages.Therefore,the fusion of two depth acquisition techniques to obtain depth information has become an important research direction in the field of three-dimensional imaging.This paper aims to improve the accuracy and robustness of deep acquisition techniques.,the ToF-binocular fusion algorithm is analyzed and studied according to the characteristics of active pulse ToF depth acquisition technology and passive binocular stereo vision depth acquisition technology.The main content and innovation of this thesis are reflected in the following two aspects:1.We propose a pulse ToF depth denoising algorithm based on semi-blind deconvolution.Demand for improved depth information accuracy,the algorithm directly to the design depth of the pulse ToF camera original measurements of regularization,avoid the traditional method may cause error to pixel value itself,combined with the total second-order generalized variational method to optimize the objective function,the algorithm firstly estimate the point spread function(PSF)of each pixel point,makes the optimization goal into a non blind deconvolution problem.Secondly,the algorithm makes good use of the advantage of the second order TGV model to balance the first and second derivatives adaptively,and effectively retains the edge information of depth image while denoising.The design of the denoising algorithm not only improves the quality of the ToF depth map,but also effectively improves the accuracy of the subsequent fusion algorithm,laying the foundation for the ToF-binary fusion algorithm.2.We propose a depth acquisition algorithm based on the fusion of pulse ToF and stereo matching.The algorithm uses a reliable image and ToF depth modulation intensity map information to design the matching cost function,makes the stereomatching algorithm in weak texture area,and repeat area towards the depth of ToF value space trusted to find matching points,and also puts forward the matching cost aggregation algorithm based on K nearest neighbor is made to further improve the accuracy of stereo matching algorithm.In other areas,we combined the ToF depth image and the credibility function of the optimized stereo matching depth map,respectively designed the weight function based on the credibility function,and adopted a pixel-level fusion algorithm based on the weight,so as to better obtain the depth image of the scene with high precision.