Study On Depth Acquisition Method Of Structured Light Based For Dynamic Scene

In an era of increasingly demanding perception and understanding of the surrounding scene,deep acquisition technology has become the focus of extensive attention,such as automatic navigation,virtual reality,human-computer interaction and so on.Among the existing depth acquisition methods,structured light method is simple,efficient and high precision,which leads to be the focus of this paper.In this paper,we focus on the structured light with singleshot,which can achieve depth measurement with high precision and high resolution in dynamic scenes.Frist,we propose a dense single-shot color binary fringe pattern.To avoid the problem,which non-sinusoidal image acquired by camera after sinusoidal pattern is projected by digital projector,the pattern takes binary fringe as the basic element and uses digital filtering method to process projected binary fringe to obtain sinusoidal information.Then,a shorttime Fourier transform is used to estimate the truncated phase locally,and the De sequence represented the number of phase periods can be extracted with the color features to complete the phase unwrapping.In this way,the two disparity maps are aggregated and optimized to reduce the ripple effect caused by local phase estimation.Finally,by the quantitative and qualitative analysis of the target scene,the validity of this method is verified.However,due to the limitation of the continuous coding pattern,the above pattern designed by us is prone to local phase estimation errors.To improve the robustness,a discrete color grid coding pattern is proposed.During the design process,the robust coding table of high Hamming distance is adopted,and a clever feature design increases the number of symbols and reduces the size of decoding window.Meanwhile,considering the local rotation invariance at the rhombus intersection position,it is also taken as the coding feature to increase the number of point clouds,which improves the density of the depth data obtained by this sparse coding method.Experimental data show that although the number of point clouds obtained by this method is still sparse compared with the previous pattern,its robustness and the measurement accuracy is further improved.