基于曲率尺度空间的激光拼焊焊缝边界点识别方法
张万江; 许敏
刊名焊接学报
2013
卷号34期号:7页码:93-96, 117-118
关键词激光拼焊 结构光 特征点识别 曲率尺度空间
ISSN号0253-360X
其他题名Curvature scale space-based boundary points detection of weld seam of tailored blanks laser welding
产权排序2
中文摘要针对激光拼焊焊缝结构光光纹图像畸变小、特征点不明显且易受噪声干扰的特点,提出了一种基于改进的曲率尺度空间的特征点识别方法.该方法首先采用较高尺度的高斯卷积模板,对光纹曲线进行卷积滤波并使用自适应K-余弦法计算轮廓上每一点的曲率并选取特征点;然后在小尺度高斯模板下跟踪定位大尺度模板下获得的特征点候选点,以获得小尺度模板下光纹曲线特征点的准确位置.在光纹曲线的离散曲率计算过程中,考虑了支持区域对曲率计算的影响,提出了一种自适应K-余弦法,与传统的离散曲率计算公式相比,自适应K-余弦法具有更好的抗干扰能力.
英文摘要Feature point detection of laser stripe is the key technology for automatic seam quality inspection. Due to small aberrance and sensitive to noise of laser stripe for tailored blanks laser welding, the traditional feature point extraction methods are difficult to obtain accurate feature points. On the basis of analyzing the image features of laser stripe,an improved multi-scale feature point detection method was proposed based on curvature scale space ( CSS) technique. Firstly,a small-scale Gaussian template is used to smooth the centerline of laser stripe. Secondly,the improved CSS adopts an adaptive K-cosine algorithm,which has a dynamic region of support and excellent noise suppression performance to calculate the curvature of centerline smoothed by large-scale Gaussian template. Thirdly,the local extrema of the curvature extracted by a threshold T are used to candidate feature points. Lastly,the candidate feature points are refined in small-scale Gaussian template. The results showed that the improved CSS algorithm has better robust and anti-interference than the that of CSS algorithm.
收录类别EI ; CSCD
语种中文
CSCD记录号CSCD:4903043
公开日期2013-12-26
内容类型期刊论文
源URL[http://ir.sia.cn/handle/173321/13938]  
专题沈阳自动化研究所_智能检测与装备研究室
推荐引用方式
GB/T 7714
张万江,许敏. 基于曲率尺度空间的激光拼焊焊缝边界点识别方法[J]. 焊接学报,2013,34(7):93-96, 117-118.
APA 张万江,&许敏.(2013).基于曲率尺度空间的激光拼焊焊缝边界点识别方法.焊接学报,34(7),93-96, 117-118.
MLA 张万江,et al."基于曲率尺度空间的激光拼焊焊缝边界点识别方法".焊接学报 34.7(2013):93-96, 117-118.
个性服务
查看访问统计
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。


©版权所有 ©2017 CSpace - Powered by CSpace