基于三维测距的坝面施工人机碰撞预警方法

A human-machine collision warning method for dam surface construction based on three-dimensional ranging

摘要

【目的】在混凝土坝建设过程中，立体交叉作业及受限的施工区域导致安全风险加剧，传统二维监控缺乏深度信息，难以精准判断人员与机械的相对位置，施工安全预警效果有限。为解决单目视觉测距尺度不确定性问题，提升坝面施工的人机碰撞风险识别与实时预警能力，提出一种基于单目视觉深度感知的施工安全预警方法。【方法】构建了YOLOv10目标检测与MiDaS单目深度估计融合的三维测距模型，通过目标识别、深度估计与空间坐标转换实现三维欧氏距离的实时测算，并结合深度图与安全阈值建立时空联合判据的预警机制。为验证方法有效性，设计了与主流单目深度估计模型的对比试验，并依托四川某水电工程开展实例测试。【结果】试验结果显示：该模型在混凝土吊罐离地高度测量中的均方根误差为27 cm,在人机相对距离检测中的均方根误差为35 cm,单帧处理耗时0.05 s,能够满足坝面施工现场实时监测的需求。与传统二维视频方法相比，该模型在风险识别的准确性与实时性方面均表现出明显优势。【结论】研究成果构建了基于三维测距的施工安全预警方法，实现了混凝土坝面施工的立体化风险监测与危险预警。不仅可服务于坝面施工场景，还具备拓展至人机协同作业与施工数字孪生系统的潜力，为未来实现更高水平的多目标预警与动态防护提供参考。

Abstract

[Objective] During the construction of concrete dams, the increased safety risks due to three-dimensional intersecting operations and limited construction areas, along with the lack of depth information in traditional two-dimensional monitoring, make it difficult to accurately determine the relative positions of personnel and machinery. Consequently, the effectiveness of construction safety early warnings is limited. To address the scale uncertainty in monocular vision-based ranging and enhance the human-machine collision risk identification and real-time warning capabilities for dam surface construction, a construction safety early warning method based on monocular visual depth perception is proposed. [Methods] A three-dimensional ranging model was constructed by integrating YOLOv10 target detection with MiDaS monocular depth estimation. Real-time calculation of the 3D Euclidean distance was achieved through target identification, depth estimation, and spatial coordinate conversion, and a warning mechanism based on spatiotemporal joint criteria was established by combining depth maps and safety thresholds. To verify the effectiveness of the method, comparative experiments with mainstream monocular depth estimation models were designed, and case tests were conducted based on a hydropower engineering project in Sichuan. [Results] The experimental result showed that the model achieved a root mean square error(RMSE) of 27 cm in measuring the height of concrete buckets and 35 cm in detecting the relative distance between humans and machines. The processing time per frame was 0.05 s, meeting the requirements for real-time monitoring at dam surface construction sites. Compared to traditional two-dimensional video method, the model demonstrated significant advantages in both the accuracy and real-time performance of risk identification. [Conclusion] A construction safety early warning method based on three-dimensional ranging is established, enabling three-dimensional risk monitoring and hazard warning for concrete dam surface construction. This method not only serves dam surface construction scenarios but also has the potential to be extended to human-machine coordinated operations and construction digital twin systems, providing a reference for achieving higher levels of multi-target early warning and dynamic protection in the future.

关键词

混凝土坝 / 立体交叉作业 / 碰撞预警 / 三维测距 / 单目深度估计 / 风险评估 / 施工安全预警 / 施工数字孪生系统

Key words

concrete dam / three-dimensional overlapping operations / collision warning / three-dimensional ranging / monocular depth estimation / risk assessment / construction safety warning / construction digital twin system

基金资助

国家自然科学基金项目(52209163)