实木板材无损检测中三维点云数据的预处理

在实木板材检测过程中,利用激光线扫描技术,以设定的速度扫描待测木材的表面,获取三维数据,可实现实木板材无损检测、缺陷定位和三维模型重构。然而,三维激光轮廓扫描仪采集的原始数据不仅存在噪声的干扰,而且其密度一般也非常大,影响检测结果和后续重构。通过对采集到的实木板材表面轮廓点云数据进行预处理,不仅可以去除噪声干扰,降低点云密度,而且可将对后续重构无影响的背景点云数据去除,简化重构过程,保证曲面重构过程的精度、简度和速度。使用Chroma+Scan3350型激光轮廓扫描仪,配合实木板材无损检测装置,采集了赤松板材和樟子松板材图像,围绕采集到的原始点云数据进行预处理。在分析点云数据的噪声来源的基础上,对噪声数据进行分类,并依次滤除,可通过曲线检查法和弦高差法去除少量的异常点,根据线扫描点云数据特点,在对比高斯滤波、平均滤波、中值滤波和小波滤波的平滑处理结果后,最终选取小波滤波进行数据平滑,取得了良好的滤波效果,很好地保留了原始点云数据的上升沿、下降沿等特征信息和曲线的棱角特点。对比角度-弦高联合法与曲率采样精简法的点云数据精简结果,选取曲率采样精简法对数据进行精简,精简效果好,速度较快。
In recent years, 3D measurement technology has been developed rapidly, and widely used in many fields, such as surveying and reverse engineering. In the process of solid wood panel nondestructive testing, surfaces of solid wood panels could be scanned at a set of constant speed using the laser scanning technology to obtain the three-dimensional data. In this way, the nondestructive testing of solid wood panels was conducted, the defects were detected and 3D models were reconstructed. However, there was much noise in the original point cloud data acquired by the 3D laser scanner, and the density of the point cloud data was high, which affected detection results and subsequent reconstruction. Pre-processing of point cloud data collected from solid wood panels could denoise and decrease density of point cloud data. While the background point cloud data, which were not used for reconstruction, should be also removed to simplify the reconfiguration process, ensure the accuracy, simplicity and speed of reconstruction. Acquired by the Chroma+Scan3350 laser scanners with the nondestructive device, point cloud data of the original solid wood panels, such as Pinus densiflora and P. sylvestris var mongolica were pre-processed. Based on the analysis of the noise sources of the point cloud data, the noise data were classified and filtered. Curve check method and chord deviation method were used to remove the outliers. According to the characteristics of the line scanning point cloud data, a variety of point cloud was smoothened and point cloud simplification methods were used. Compared the gauss filtering, mean filtering, median filtering and wavelet filtering, the wavelet filter was finally selected for data smoothing and we achieved a great filtering outcomes. This approach could preserve angular features of curve by preserving rising edge and falling edge of original point cloud data. Comparing the angle-chord deviation method with the curvature sampling streamline method,we finally selected the curvature sampling to simplify the data. The curvature sampling