The Equipment Detection And Localization Of Large-Scale Construction Jobsite By Far-Field Construction Surveillance Video Based On Improving Yolov3 And Grey Wolf Optimizer Improving Extreme Learning Machine

The equipment detection and localization on large-scale construction jobsite by far-field construction surveillance video is significant to monitor jobsite safety and construction process. These videos usually with a larger view containing extensive construction equipment and richer environment information could help the supervisor know more about real-time status of the large-scale construction jobsite. However, due to numerous multi-scale objects, especially for small scale objects in the video, it is difficult to recognize each equipment intuitively, which is not conducive to ensuring jobsite security and guiding construction process. This study presents a real-time multi-scale equipment detection and localization model including the improving YOLOv3 for detection and the grey wolf optimizer improving extreme learning machine (GWO-ELM) for localization. The effectiveness and efficiency of the proposed model are demonstrated via a real hydraulic-engineering megaproject. The experimental results show that the improving YOLOv3 has improved 4.8% mean average precision (mAP) over the multi-scale objects detection, especially, the significant improvements of 17.2% mAP and 13.3% recall at small scale objects detection while maintaining real-time. The improving YOLOv3 is proved robust under various environmental conditions on construction monitoring. Meanwhile, the mean absolute localization percentage errors of different scale objects are less than 1%, which indicates the accuracy and effectiveness of the model to assist supervisor in inspecting construction safety and process. (C) 2021 Elsevier Ltd. All rights reserved.