Research Papers

Stress monitoring using wearable sensors for pedestrian worker risk analysis

Version 1
Date added July 10, 2018
Downloaded 0 times/fois
Category 2018 CARSP XXVIII Victoria
Tags Research and Evaluation, Session 6B
Author/Auteur Lafond, Vachon, Gagnon, Godillon, Cloutier
Stream/Volet Research and Evaluation
Slidedeck Presentation Only (no paper submitted)

6B - Cloutier


Pedestrian road workers are particularly vulnerable when it comes to workplace accidents. These workers are exposed daily not only to road accident risks, but also to levels of stress and insecurity that may be indicative of potential occupational injuries. Measuring stress variations during work activities may help pinpoint situations that pose greater risks and allow taking better preventive measures. Here we report initial results from an ongoing field study investigating the feasibility of detecting meaningful stress variations of police officers performing traffic control duties. The study seeks to collect wearable sensor data to calibrate a stress assessment model for this type of context and perform initial contextual analyses to determine factors impacting stress levels. Sixteen police officers were monitored during different traffic control periods and at different locations in Quebec City and Montreal, Canada, for a total of 5 hours each on average. Workers were required to wear a Zephyr Bioharness 3 chest strap measuring cardiac and respiratory activity. Subjective stress was measured on a 10-point scale at the start and end of the whole work session. Change in subjective stress (same/increase/decrease) was also measured at each 15-minute interval. Observers took note of these stress ratings, recorded occurrences of potentially stressful traffic incidents, and recorded several characteristics linked to current activity (on-task or off-task), presence of road construction sites, number of lanes in construction and the (low/high) level of traffic (cars and pedestrians) on site. The unit of analysis selected was each 15-minute period ending with a subjective stress measure (n = 315). Metrics of cardiac and respiratory activity (averages and deviations) were calculated for each interval. A subjective stress metric was used to calculate the cumulative perceived stress for each interval going from the initial anchor (0-10), to higher or lower values over time (-1/+1 for each decrease/increase). To reduce noise in the data, stress levels were divided in two distinct states (low vs high) based on a clear bimodal distribution of the data. Binomial logistic regression analyses showed that these two states were significantly associated to current activity (on-task vs. off-task; p = .012), road traffic level (p = .044), pedestrian traffic level (p < .001), presence of road construction (p = .041), and observed occurrence of a stressful incident (p < .001). A multifactorial logistic regression was used to predict subjective stress from psychophysiological signal data, achieving a predictive accuracy of 67% on cross-validation tests. Stress measurements were indeed sensitive to potential risk factors in the context of pedestrian workers. The psychophysiological stress model could be improved to better take into account individual differences when a larger dataset becomes available and eventually eliminate the need for subjective measures. This feasibility study demonstrated the potential of objective, physiological stress measurement for occupational road injury prevention. Directions for future work include a more extensive data collection, measurement of electrodermal activity and analysis of video data for more fine-grained traffic event-based analyses.