Research On Background And Highlight In 3D Dynamic Fringe Projection

Fringe projection 3D shape measurement is a commonly used 3D measurement technology and it can quickly and accurately obtain the 3D shape of the object.After capturing the deformed fringe pattern,it analyzes and extracts the effective information to obtain the threedimensional information of the object.However,in practical applications,the environment and objects are various,so that the process of measurement is affected by interference factors such as uneven background light and highlight reflection,resulting in large errors in the results of measurement.It becomes tricky in dynamic measurement due to one or rare fringe patterns.Therefore,on the premise of dynamic measurement,this paper conducts an in-depth study on the above problems.The contents and innovations of the research are as follows:1)In order to improve the efficiency and effect of the background light elimination from a single fringe pattern,a fast and adaptive bidimensional sinusoids-assisted empirical mode decomposition algorithm is proposed.The parallel calculation of the proposed algorithm is also implemented.First,the theoretical process of the empirical mode decomposition method is deeply analyzed.The main factors that determine the efficiency and effect of the method are determined,which are the size of the morphological operator,the size of the convolution kernel and the size of the auxiliary sinusoids.The period of the main fringe components in each iteration is analyzed and calculated.The period is set as the size of the above operation operator,which greatly reduces the number of iterations and avoids errors caused by excessive decomposition.The size of the morphological operator and the size of the convolution kernel are decomposed to achieve parallel acceleration.2)In order to repair the loss of local fringe caused by high reflection in measurement,a deep learning-based method for inpainting the highlight region of fringe pattern is proposed.First,a low-darkness binary grid pattern is designed based on the projected sinusoidal fringe pattern,which is dominated by the valley of sinusoidal signals.Then,the binary grid pattern is projected onto the surface of the object,and the grids are modulated by the object.The modulated information is the deformation corresponding to the lost fringes in the highreflection fringe pattern.Finally,the network of deep learning is trained to generate sinusoidal fringes with the deformed grid lines as a priori to repair and supplement the lost fringes.This method does not require to adjustment the hardware system.It only needs to project a binary grid pattern to process the objects with high dynamic range.In addition,this paper uses graphics software to construct a virtual system,which solves the problem of data acquisition of deep learning training sets.A large number of experiments have proved that the proposed method is fast,effective and has good generalization.