Abstract: The reliability of tower crane wire rope is directly related to the efficiency of the construction site and the safety of workers. In view of the existing problems of focusing on damage type identification while neglecting rope diameter measurement, leading to insufficient reliability and efficiency of detection, the study proposes a method of wire rope measurement and damage identification integrating binocular stereo vision imaging technology and a deep learning model. Firstly, a binocular stereo vision imaging device is installed on the crane wall of the tower crane, employing an NVIDIA B01 IMX209 binocular stereo vision camera to capture the rope damage images containing depth information. Subsequently, the depth information clustering method is used to carry out the precise segmentation of the wire rope, and the similar triangle principle was used to complete the rope diameter measurement and abnormal diagnosis. Finally, the segmented image based on depth information clustering is used to train the ResNet-50 model to diagnose the specific damage types of steel wire rope. The results show that the average accuracy of the proposed method is 94.64%, which is 15.01%, 28.06% and 18.73% higher than that of YOLOv8n, MobileNetV3-v8, and GhostNet-v models, respectively.

Key words: binocular stereo vision, ResNet-50, similar triangles, depth information