Autonomous bolt tightening is a challenging task for maintenance robot on power transmission lines, because bolts could hardly be traced and recognized without manual command. In this paper, a structure of bolt tightening robot and its bolt tracing, recognition and positioning strategy utilizing vision detection are proposed. The bolt tightening robot is equipped with a camerainstalled bolt tightening unit which is connected by an arm with three joints, the proposed bolt detecting strategy consists of two steps. First is bolt tracing in which the drainage wire is used as a reference, through the visual detection of the location and direction of the drainage wire, the bolt tracing task can be carried out by tracing along the wire, thus the tracing process is simplified and the difficulty of vision recognition in complicated environment is reduced. The second step is bolt recognizing step in which an improved Hough transform is proposed and the center of the circle shape edge is utilized to verifying the recognizing result. To make the bolt detecting more reliable, an initial classification algorithm utilizing HOG and SVM techniques is applied at the very beginning. The experimental results show that the proposed strategy can detect the bolt efficiently and pave the way for robotbased automatic bolt tightening live work on lines.