基于概率连续模型的激光视觉焊缝自动跟踪

Laser Vision Seam Automatic Tracking Based on Probability Continuous Model

针对目前在实际焊接过程中多数焊缝自动跟踪系统对飞溅、弧光等噪声干扰十分敏感,从而造成焊接精度损失的问题,设计了能够实时检测焊缝特征图像的线激光视觉传感器,并根据其几何模型建立了精确的测量模型。跟踪过程中以线性表示模型对观测矢量进行建模并利用仿射变换模型对焊缝运动进行描述,提出了基于概率连续模型的焊缝跟踪算法。充分利用图像中激光条纹和背景噪声的空间一致性,结合刻画邻域结构内像素点间相互关系的一阶马尔可夫随机场理论,推导出焊缝跟踪问题的目标函数。采用基于最小二乘法与最大流/最小割的迭代算法对其进行求解,最终获取准确的焊缝位置。搭建了焊缝跟踪试验平台,并在实际焊接环境中应用所提算法进行焊缝跟踪试验。试验结果表明该算法的跟踪精度达?0.109 1 mm,平均每帧图像处理时间不长于45 ms,并且激光条纹与焊接熔池的最小距离可达24 mm,能够克服强烈噪声干扰,实现实时、准确的焊缝跟踪。

Aimed at the current problem that the majority of seam tracking systems are very sensitive to splash, arc and other noise in the actual welding process, which result in loss of bonding accuracy, a line laser vision sensor which could detect characteristic images of weld seam in real time is designed, and an accurate measurement model based on its geometry model is built. By modeling the observation vectors using a linear representation model and using the affine transformation model to describe the motion of the weld in tracking process, a seam tracking algorithm based on the probability continuous model is proposed. To derive the objective function of seam tracking problem by taking advantage of the spatial consistency of laser stripe and background noise in the captured images and combining the first order Markov random field theory which reflects the relationship between the pixels in the neighborhood structure. Solve this function by iterative algorithm based on the least squares and max flow/min cut, and get the exact location of the weld ultimately. Built an experiment platform and apply the proposed algorithm to the welding seam automatic tracking experiment in the actual welding environment. Experimental results show that tracking precision is up to0.109 1 mm, the average processing time for each frame45 ms of this algorithm and the minimum distance between the laser stripe and weld pool is up to 24 mm, which prove that the proposed algorithm can overcome the strong noise and achieve real-time as well as accurate seam tracking.