基于局部特征重识别的多台臂动施工机械工效分析方法-《建筑科学与工程学报》

文章信息/Info

Title:: Multiple articulated construction machinery efficiency analysis method based on local feature reidentification

作者:: 卢昱杰，钟利剑，魏伟，王硕; 1. 同济大学土木工程学院，上海 200092； 2. 同济大学上海自主智能无人系统科学中心，上海 201210

Author(s):: LU Yujie, ZHONG Lijian, WEI Wei, WANG Shuo; 1. College of Civil Engineering, Tongji University, Shanghai 200092, China； 2. Shanghai Research Institute for Intelligent Autonomous System, Tongji University, Shanghai 201210, China

Keywords:: multiobject re-identification; local feature matching; multi-object tracking; machinery efficiency analysis; construction machinery management

摘要:: 为克服现有基于视觉的机械工效分析方法在应用于臂动施工机械时易受遮挡影响而跟踪中断的问题，从而实现长时间、智能的施工机械工效分析，提出一种基于局部特征重识别的多台臂动施工机械工效分析方法。该方法利用卷积神经网络提取臂动施工机械的局部外观特征，将其用于身份匹配与目标重识别，可有效纠正因遮挡导致的多目标身份误判错误，在多目标跟踪的基础上，进一步应用活动识别算法实现了目标活动统计与工效分析。以上海某施工项目为研究案例，对其施工现场的挖掘机进行了跟踪和工效分析。结果表明：采用该方法的机械重识别与稳定跟踪（视频时间25 min）结果较为精确，机械工效分析精度达到96.74%；该研究为施工领域的多机械目标重识别提供研究范式，为基于视觉方法的复杂场景机械工效智能分析提供实践参考。

Abstract:: In order to overcome the interruption issues caused by occlusion in existing vision-based machinery efficiency analysis methods applied to articulated construction machinery, and to achieve longterm and intelligent efficient construction machinery efficiency analysis, a method for efficiency analysis of multiple articulated construction machinery based on local feature reidentification was proposed. The proposed method utilized convolutional neural networks to extract local appearance features of the excavator for identity matching and target re-identification, effectively correcting the misjudgment of multiple target identities caused by occlusion. Based on multiobject tracking, the activity recognition algorithms to achieve statistical analysis of multiple excavators activities and efficiency analysis were further applied. Taking a construction project in Shanghai as a case study, the excavator on the construction site was tracked and the efficiency was analyzed. The results show that the results of reidentification and stable tracking (with a video duration of 25 minutes) using this method are more accurate, with the accuracy of machinery efficiency analysis reaching 96.74%. This study provides a research paradigm for multi-mechanical target re-identification in the construction engineering field and offers a practical example for intelligent efficiency analysis based on visual methods in complex scenarios.