Head-Integrated Detecting Method for Workers under Complex Construction Scenarios

Abstract

Real-time detection of workers is crucial in construction safety management. Deep learning-based detecting methods are valuable, but always challenged by the possibility of target missing or identity errors under complex scenarios. To address these limitations, previous research depended on re-training for new models or datasets, which are prohibitively time-consuming and incur high computing demands. However, we demonstrate that the better detecting model might not rely on more re-training of weights; instead, a training-free model can achieve even better performance by integrating head information. In this paper, a new head-detecting branch (55 MB) is added to the Keypoint Region-based Convolutional Network (Keypoint R-CNN, 226 MB) without altering its original weights, allowing for a less occluded head to aid in body detection. We also deployed motion information and anthropometric data through a post-processing module to calculate movement relationships. This study achieved an identity F1-score (IDF1) of 97.609%, recall (Rcll) of 98.173%, precision (Prcn) of 97.052%, and accuracy of 95.329% as a state-of-the-art (SOTA) method for worker detection. This exploration breaks the inertial attitudes of re-training dependency and accelerates the application of universal models, in addition to reducing the computational difficulty for most construction sites, especially in scenarios with an insufficient graphics processing unit (GPU). More importantly, this study can address occlusion challenges effectively in the worker detection field, making it of practical significance.

Bidirectional Tracking Method for Construction Workers in Dealing with Identity Errors

“…As depicted in Figure 2, the proposed method consists of the detector, imputers, refiners and tracker; among them, the first three items are detailed and discussed in [10] while the tracker is our concern in this paper. Whereas the unidirectional method only contains the body_tracker, our tracker contains four parts:…”

“…Motion data, including position and velocity, can be effectively extracted using a linear Kalman filter (KF) [10]. The KF estimates the state of a dynamic system from noisy measurements, employing a state equation and an observation equation, as depicted in Equations ( 1) and (2).…”

“…In summary, the intra-frame processing stage employs a combination of confidence scores, size principles and keypoint analysis to refine head and body detections, reducing false positives and preparing the data for more accurate inter-frame tracking using the RseNet-50-based [10] head-integrated keypoint R-CNN model.…”

“…(3) Worker tracking, unlike pedestrian tracking, encounters more occlusion challenges in complex scenarios [10], posing higher ID error risks. While pedestrian tracking algorithms typically overlook ID-related metrics, worker tracking, which is crucial for safety management, necessitates their consideration, particularly in MOT challenge-like contexts [11,12].…”

“…According to the literature [2], instance segmentation has been posited to enhance detection performance, as exemplified by Xiao et al's [2] adaptation of the Mask R-CNN [20] for worker tracking. However, their study relies on datasets without annotations for heavily occluded individuals (occlusion rate above 70%), which precludes the possibility of improving the data association process, as previously noted by [10].…”