关键词: 双目立体视觉, 测距技术, 线路外破监测, 掩膜区域卷积神经网络, 部分标定

Abstract: Aiming at the problems of inaccurate selection of invasive hazards points in monitoring images and insufficient calibration space in actual applications, this paper proposes an automatic selection algorithm for invasive hazards points and improves the calibration process of binocular stereo vision. Practical application in the measurement of the invasive hazards distance of the corridor in the complex environment. The article first summarizes the binocular stereo vision algorithm for the risk measurement of invasive hazards in large scenes under ideal conditions. Then, Mask RCNN is used to identify the outer broken contour and the dangerous point closest to the line. At the same time, for the possible complex terrain, a partial calibration method to reduce the calibration area is proposed. Finally, some typical external breach intrusion scenarios are selected to verify the effectiveness of the dangerous point extraction algorithm and calibration method. The results show that the improved algorithm keeps the identification accuracy of dangerous points below 0. 2 m when monitoring the external damage of construction machinery, and can make the binocular distance measurement applied to the task of monitoring external damage of the line in the complex environment.

Key words: binocular stereo vision, distance measurement, transmission line monitoring, mask region convolutional neural network( Mask-RCNN), partial calibration