Sensing Workers Gait Abnormality for Safety Hazard Identification

Abstract

-Ironwork is considered one of the most dangerous construction trades due to its fall-prone working environment. Since safety-hazard identification is fundamental to preventing ironworkers' fall accidents, engineering measures have been applied to eliminate fall hazards or to reduce their associated risks. However, a significant quantity of hazards usually remains unidentified or not well assessed because most current efforts rely on human judgment to identify hazards. To enhance hazard identification efforts, this paper develops a technique for detecting the jobsite safety hazards of ironworkers by analyzing their gait anomalies. Using wearable inertial measurement units (WIMUs) to record kinematic data about ironworkers' gait, this study collected kinematic data while the workers interacted with two types of jobsite hazards. The anomaly level of each gait was modeled using diverse gait-related metrics. Moreover, relationships between safety hazards and worker gait abnormalities were examined through extensive experiment evaluations. The results reveal opportunities for enhancing hazard identification performance by monitoring workers' bodily response.

An Efficient Automatic Gait Anomaly Detection Method Based on Semisupervised Clustering

“…e key assumption of our proposed method is that the deviation from the normal gait baseline pattern indicates a possible gait anomaly. is assumption has been verified in some earlier research studies based on both the nonmachine learning approach [50] and the machine learning approach [15]. In our observation, by representing the mean and the standard deviation values of the proportions of the silhouette images in walking sequences overlapping with the image E, the feature vectors of the abnormal gait sequences resembled each other and thereby can be classified as a single class.…”

Motion Capture Technology in Industrial Applications: A Systematic Review

“…In the pre-processing stage, data were filtered to mitigate noise and address drift, outliers and missing points in data streams (to avoid de-synchronisation for instance), then were fused together, and were further processed to extract the numerical values of interest; in the studies considered by this review, this was mostly achieved via low-pass filters (e.g., nth order Butterworth, sliding window and median filters) [ 15 , 31 , 40 , 45 , 58 , 61 , 62 , 63 , 66 , 69 , 73 , 75 , 77 ], Kalman filters [ 17 , 28 , 29 , 40 , 41 , 51 , 54 , 57 , 60 , 71 , 73 ] and band-pass filters when EMG data were collected [ 29 , 49 , 50 , 51 , 54 ]. The drift of inertial data, a typical inertial sensors issue, was sometimes addressed in the pre-processing stage by implementing filtering methods such as the zero-velocity update technique [ 44 , 59 , 60 ].…”

“…Two works also used the Oculus Rift virtual reality headset to remotely assess industrials locations and control robotic elements [ 39 , 43 ]. The tracking accuracy of the developed systems was directly assessed against gold-standard MoCap systems (e.g., Vicon or Optotrack; Table 8 , in bold) in six works [ 14 , 15 , 55 , 59 , 73 , 77 ], while the classification or identification accuracy of a process was frequently evaluated with visual inspection of video or phone cameras [ 15 , 29 , 36 , 44 , 60 , 63 , 69 ]. A thorough diagram showing the connections between type of industry, application and MoCap system, for each considered study is also presented on Figure 5 .…”

“…When MoCap systems were combined with EMG sensors ( Table 8 ), the musculoskeletal and postural evaluation of workers during generic physical activities ( Table 5 ) was improved [ 29 , 48 , 49 , 50 , 51 , 54 , 57 ]. Inertial sensors also showed good results for the identification of hazardous events such as trips and falls in the construction industry [ 44 , 58 , 60 , 65 , 66 , 69 , 75 ], but the positions and numbers of the used IMUs were reported to impact on the intra-subject activity identification [ 26 ]. For example, fewer IMUs placed on specific anatomical sections (e.g., hip and neck) showed similar task classification performance than a greater number of IMUs distributed on the entire body [ 36 ].…”