基于Faster RCNN的双目视觉焊缝匹配研究

Study on Seam Matching of Binocular Vision Based on Faster RCNN

针对目前第三代自主焊接机械手在路径规划中提取焊缝特征所遇到的噪声问题,提出了一种基于Faster RCNN的快速目标区域提取算法。使用Faster RCNN算法锁定焊缝特征所在区域,去除噪声特征,再使用Sojka特征点提取特征,最后完成匹配,为后续三维重建做准备。结果表明,该方法缩小了特征点的提取范围,有效降低了由于目标环境或自身结构所带来的干扰,能够快速去除影响焊缝特征三维重建非焊缝区域的特征点。

Aiming at the noise problem of the third generation autonomous welding manipulator in the path planning to extract weld features,a fast target region extraction algorithm based on Faster RCNN is proposed.The Faster RCNN algorithm is used to lock the area where the weld features are located,remove the noise features,and then use Sojka feature points to extract the features.Finally,the matching is completed to prepare for the subsequent three-dimensional reconstruction.The results show that the method reduces the extraction range of feature points,effectively reduces the interference caused by the target environment or its own structure,and can quickly remove the feature points that affect the weld feature 3D reconstruction of non-weld zone.