Abstract: This paper uses structure laser as the active light source. A narrowband filter weakens the arc light, which can improve SNR（Signal Noise Ratio） of image effectively. Experimental image only extracts the red component. The method strengthens the specific wavelengths and weakens others. This paper introduces an improved median filtering algorithm. The algorithm has three advantages. It effectively improves the performance of noise reduction. It retains the details of the image. The complexity of the algorithm is about[O(N)]. The weak noise is eliminated by binarization. At the same time, the laser stripe is reserved. This paper adopts a multi-scale edge detection approach. This approach can better extract the complex image edge. When recognizing characteristic quantity, it can enhance fault-tolerant. When extracting location information of the horizontal section by black pixel accumulation algorithm, it can fix location for the stripes. The system can identify the weld information accurately and extract enough information, completing the weld tracking.

Key words: image processing, image segmentation, edge extraction, feature point extraction

摘要： 采用激光线结构光作为主动光源，用滤光片滤掉弧光，有效提高了图像的信噪比得到清晰的焊缝图像，而且实验图像仅提取红色分量，进一步把特定波长的光放大，削弱其他波长的光。采用一种改进的中值滤波，此方法使降噪性能有效提高，较好保留图像细节，算法复杂度近似为[O(N)]。通过二值化把信号不强的噪声消除，同时保留激光条纹信息。选用一种多尺度边缘检测方法，它能较好地提取复杂的图像边缘。提取特征点时，进一步加强系统的容错能力，提取水平段位置信息时，采用黑色像素行累加值的办法，准确定位条纹位置。系统准确地识别焊缝特征并提取足够的信息，将焊缝的图像坐标转换成世界坐标，完成跟踪焊缝。

关键词: 图像处理, 图像分割, 边缘提取, 特征点提取