Depth Acquisition By Combining Structured Light And Stereo Matching

With the improvement of the society and our living standards,the 3D information is more and more widespread in the industry and our daily life,such as virtual reality,reverse engineering and unmanned driving etc.Meanwhile,the study of depth acquisition becomes a hot topic for researchers.In this paper,we carried out research into existing depth acquisition technologies and proposed two methods in which structured light and stereo matching methods are combined.The first method is based on the three-step phase shifting structured light method.The depth acquisition system contains one projector and two cameras with exact same intrinsic parameters.These three devices are mounted in the same horizontal plane with the optical axis parallel.After projecting the three patterns sequentially,two cameras capture the deformed patterns.Then,the phase data is calculated and the initial scene image is recovered.To improve the accuracy and minimize the matching error,the belief propagation method is adopted to correspond the recovered two scene images in this paper.Then,the coarse depth obtained from stereo matching is used to benefit the phase unwrapping.The high accurate depth data is achieved in the end.In this method,both the active phase-based structured light and the passive stereo matching are combined to acquire the depth.So the method achieves high accuracy and is robust to the complex scenes,especially when the scenes have rich texture,color and depth discontinuities.However,the previous method needs three patterns to acquire depth.this method might not be suitable for dynamic scenes.We proposed another method in which only one sinusoidal fringe pattern is adopted to acquire high accurate depth data.The depth acquisition system of this method is the same as that of the first method,In this method,only one pattern is projected.The Gabor filter is adopted to calculate the wrapped phase from the deformed pattern which is captured by the two cameras.At the same time,we recover the initial scene by spectral filtering.Like the first method,we acquire the coarse depth by stereo matching using the belief propagation,then the coarse depth is used to do phase unwrapping.The feasibility and robustness of the method were validated by experiments.Because only one pattern is required in the second method,it can be used in dynamic scenes.