提出一种基于关键点训练学习的防振锤滑移检测方法。首先利用改进SSD模型对防振锤进行识别、定位;再选择防振锤关键点,训练MobileNetV3网络,通过上一级定位结果设定输入区域,从而实现关键点的检测;最后依据线路图像特征,制定相应判别规则。当档距内悬挂m(m≥2)个防振锤时,利用关键点几何约束关系实现判别;当档距内悬挂单防振锤时,采用EPnP算法估计相机在多个角度的位姿,通过位姿与线夹关键点像素坐标间的关系求解最邻近点的世界坐标,判断最邻近点与防振锤间的距离是否在阈值范围内。实验结果表明,所提方法能对滑移异常进行有效识别,为输电线路异常检测提供了新思路。

This study proposes a method for detecting slipping of stockbridge dampers based on key point training and learning. First, an improved SSD model is used to identify and locate the stockbridge damper. Thereafter, the key points of the stockbridge damper are selected, the MobileNetV3 network is trained, and the input area is set by the upper positioning results of the stockbridge damper to realize the detection of key points. Finally, discrimination rules are formulated according to the characteristics of line images. For m(m≥2) stockbridge dampers, the geometric constraint relationship among the key points is used to realize determination. For a single stockbridge damper, the EPnP algorithm is used to estimate the multiangle pose of the camera. Moreover, the spatial coordinates of the nearest points are obtained from the relationship between the pose and the pixel coordinates of the key point of the damp to determine whether the distance between the nearest points and the stockbridge damper is within the threshold range. The experimental results show that the proposed method can effectively identify slip faults and provide new ideas for detecting defects in transmission lines.