For the thin steel plate butt welding of containers, a visual control method of welding robot is proposed to recognize starting position and drive the weld torch to accurately track the starting position. Firstly, an iterative circle Hough transformation is presented, which can automatically obtain the region of interest including the torch characteristic point. In the second iteration, the torch characteristic point is recognized and extracted, robustly and quickly. Then the seam is recognized and its characteristic point is extracted by using line Hough transformation. Furthermore, an image space based visual control scheme of starting position is designed, and the accurate starting position control is achieved. The experimental results show that the proposed scheme can extract the torch's and seam's characteristic point coordinates, accurately and robustly. Moreover, the torch can be driven to the starting position quickly and accurately by using the visual control. Finally, test results in the container factory also verify the effectiveness of these methods, and the satisfactory results are obtained.