Abstract: A dynamic identification method for roof cracks is proposed, which integrates YOLOv8 deep learning architecture and 3D laser scanning point cloud data, to address the problems of diverse shapes and weak characteristics of roof cracks in low light, high dust, and complex spatial structure environments in coal mines, as well as low recognition accuracy and poor stability caused by environmental noise interference. By constructing a local coordinate system for the tunnel, spatial registration and feature fusion of 3D point clouds and image data can be achieved. Using reflection component reconstruction and adaptive median filtering to enhance the visual saliency of crack areas, effectively suppressing uneven lighting and dust noise. Propose outlier removal and multi temporal point cloud registration strategies based on nearest neighbor distance statistics to eliminate recognition bias caused by device pose changes. Further introduction of benchmark plane fitting and distance threshold screening mechanism to achieve preliminary localization of fracture areas and efficient input of YOLOv8 recognition framework. By optimizing the boundary box regression process through the CIoU loss function, the accuracy of identifying the position and shape of cracks can be improved. The experimental results show that the accuracy of crack recognition by this method in typical tunnel scenes reaches 96.7%, with an average processing delay of 0.8 s. Under extremely dark （illumination＜5 lux） and dense dust （visibility＜2 m） conditions, the recognition accuracy still remains at 90.2% and 88.5%, with performance degradation rates of 6.5% and 8.2%, respectively. By optimizing the CIoU loss function, the bounding box regression error is reduced to 0.05. Therefore, this method can achieve high-precision and highly robust crack recognition in various tunnel scenarios, and the recognition results highly match the actual values in multiple dimensions such as position, size, and angle, providing an efficient and reliable technical means for coal mine roof safety monitoring.