Improving safety performance in construction using visual data analytics and virtual reality

High injury rates are a prevalent issue in the global construction industry. In fact, previous efforts have established that construction workers are roughly five times more likely to be injured than their counterparts in other industries. One reason for such poor performance is that workers often fail to recognize and manage a large number of safety hazards in dynamic construction environments. When safety hazards remain unrecognized and unmanaged, workplace accidents and injuries become more likely. Therefore, improving hazard recognition and hazard management in the construction industry is fundamental to improving safety in construction workplaces. While few research efforts have focused on understanding factors that influence hazard recognition levels (e. g. , safety training, safety climate levels, etc. ) there is still an insufficient understanding of why workers fail to recognize safety hazards in construction workplaces. Such an understanding will enhance our ability to design more robust interventions to tackle the issue of poor hazard recognition. Moreover, while there have been significant technological advancements in recent years, the construction industry lags behind other industries in adopting these solutions for construction safety applications. Accordingly, the objective of the current research was (1) understand why workers fail to recognize safety hazards, and (2) develop countermeasures that target poor hazard recognition and management levels. The first objective was accomplished by examining construction hazard recognition as an everyday visual search task – similar to an individual searching for a product in a supermarket or a radiologist examining a radiograph for tissue abnormalities – as part of an experimental effort. More specifically, the research used eye-tracking technology to examine the relationship between visual search patterns adopted by workers and the resulting performance levels (i. e. , hazard recognition performance). The results revealed that several quantifiable visual search patterns are predictive of hazard recognition performance. Accordingly, the research suggested that weakness in visual search patterns during hazard recognition efforts can adversely affect hazard recognition and the resulting safety performance. The new knowledge generated from the experiment was then adopted to develop a computer-vision based algorithm to capture visual search patterns in real workplaces on a larger scale using wearable eye-tracking devices. Having accomplished the first objective, the next objective was to develop interventions that can promote hazard recognition performance. As part of this effort, two independent interventions were developed. The first intervention was a personalized training solution which leveraged 3D stereo-panoramic and virtual environment elements to offer a hyper-realistic and immersive training experience. The intervention integrated various training experiences to promote hazard recognition and management. An experimental effort focusing on the evaluation of the training suggested that the introduction of the intervention resulted in superior hazard recognition and hazard management performance. Despite the promise of the personalized training intervention, training efforts may not be sufficient to ensure the recognition of all safety hazards – particularly because of the variety of human factors that can influence performance. Therefore, the second intervention focused on developing an artificial intelligence based solution to augment human hazard recognition performance. More specifically, the system uses computer vision algorithms along with a wearable camera to localize workers and notify workers and managers when in the proximity of hazardous conditions or objects in real-time. The system was testing in indoor and outdoor environments and the results revealed an accuracy of 96% in detecting worker proximity to static and dynamic hazards. The findings and the intervention developed as part of this research effort can be leveraged to improve hazard recog ition and hazard management in the construction industry. Such efforts will be beneficial in reducing workplace injuries and improving safety performance in the construction industry.