面结构光在机检测的叶片反光抑制技术

Blade reflection suppression technology based on surface structured light on-machine detection

薄壁叶片在结构光检测过程中,由于其表面粗糙度较小,易产生强烈的反光现象,影响求解条纹相位主值,进而无法准确重构出三维点云。本文以加工过程中的叶片作为研究对象,提出一种对在机检测过程中的条纹图像进行图像增强处理的Retinex算法,恢复条纹在高反光位置的信息。首先,对薄壁叶片的反光特性进行分析,通过实验标定出最优曝光的灰度区间和理想灰度值,建立了光圈转动角度与图像平均灰度的相机响应曲线模型,调节光圈和曝光时间至最优曝光的灰度区间并以此作为检测条件。其次,基于Retinex算法处理条纹图像,通过改进的双边滤波代替常用的高斯滤波,在去除光照的同时有效保留了条纹的边缘信息。最后,对薄壁叶片进行单目结构光检测。实验结果表明,经本文算法处理后的条纹图像,通过Canny算子检测出的条纹数量最多,图像信息熵平均增长率达18.21%,解算的相位主值误差最小,利用手持式激光扫描仪检测的标准点云进行偏差分析,点云的正、负偏差分别降至0.0589 mm和-0.0590 mm,与原点云的偏差值相比分别减少了44.6%和44.1%,表面质量得到明显改善。本文提出的图像增强算法有效抑制了面结构光检测过程中的金属表面反光。

In the process of structured light detection,the thin-walled blade is easy to produce a strong reflection due to its low surface roughness,which affects the solution of the principal value of the fringe phase.As a result,it cannot accurately reconstruct the three-dimensional point cloud.In this paper,the blade in the machining process is taken as the research object,and an image enhancement process based on the Retinex algorithm is proposed to restore the information of the stripes in the position with the highest reflectivity.Firstly,the reflective characteristics of thin-walled blades are analyzed.The gray range and ideal gray value of the optimal exposure are calibrated experimentally.The camera response curve model of the aperture rotation angle and the image’s average gray level is determined,and the gray level interval of the optimal exposure is used as the detection condition by adjusting the aperture and exposure time.Secondly,the fringe image is processed based on the Retinex algorithm.The improved bilateral filter replaces the commonly used Gaussian filter,which effectively retains the edge information of the fringe while removing its illumination.Finally,monocular structured light detection is carried out on the thin-walled blade.The experimental results show that,for the fringe image processed by this proposed algorithm,the number of stripes detected by the Canny operator is the largest,the average growth rate of image information entropy is 18.21%,and the phase principal value error of the solution is the smallest.Through the deviation analysis with the standard point cloud detected by the handheld laser scanner,the positive and negative deviations of the point cloud are reduced to 0.0589 mm and-0.0590 mm,which are reduced by 44.6%and 44.1%compared with the deviation of the origin cloud,respectively,and the surface quality is significantly improved.The image enhancement algorithm proposed effectively suppresses the reflection of the metal surface in the process of surface structured light detection.