移动高光对三维场景重建的影响与消除

针对制造车间光滑地面在阳光或灯光照射下的扫描图像往往存在反射光斑,且随相机位置变化而变化,会严重干扰图像处理和利用的问题,提出了一种识别和去除高光的方法,用于改善图像处理效果和三维重建精度。基于扫描序列图像的三维场景重建流程和SURF特征原理,分析移动的高亮反光斑对图像特征点提取和匹配的影响;对于缺乏表面纹理的灰度图像,提出一种基于逐行多级阈值和动态模板高光识别与去除方法,保留非高光点的特征信息,识别与修复同步完成。实验结果表明,无高反光的前后两帧图像的匹配特征点对比有高光情况下至少高出8%,优化后的余留匹配特征点更多;扫描一段30m长的车间场景序列图像,用提出的方法修复高光后进行三维重建,点云拼接误差减小了10cm。表明移动高光对图像序列三维重建有不可忽视的影响,本文提出的方法能有效地去除高光。

Influence and elimination of moving highlight on three-dimensional scene reconstruction

The scanned image of the smooth ground in the manufacturing workshop had reflection light spot under the irradiation of sunshine or lamplight;and it was varied with the change of the camera shooting position.Based on 3d scene reconstruction procedures and SURF algorithm principle of the Kinect scanning sequence images,influences of mobile highlight reflector on extraction and matching of feature points for reference images and images which are waiting for being matched were analyzed.The experimental result shows that compared with match feature points for the frame image with high reflection,match feature points for the frame image without high reflection is at least higher 8 ％;and remaining match feature points are more after being optimized.In addition,highlight recognition and removal method should be discussed;in terms of grey level images which lack surface texture,based on progressive multilevel threshold and dynamic template,a kind of processing method was proposed;non highlight feature information was not changed;and processing time was saved;recognition and repair were synchronously completed.The experiment shows that in terms of 30m length of sequence image for scene in the workshop which is scanned by Kinect,its light should be repaired by the method proposed in the thesis;and it should be subject to three-dimensional reconstruction;point cloud registration error should be at least reduced by 1Ocm.