The unsafe actions of workers on a jobsite can result in accidents and injuries. But by measuring their behavior, it is possible to evaluate how safely individuals are carrying out their tasks. Unfortunately, to perform such monitoring manually is both time consuming and expensive.
Now, researchers led by Professor Feniosky Peña-Mora at Columbia University (New York, NY, USA; www.columbia.edu) and Professor SangHyun Lee at the University of Michigan (Ann Arbor, MI, USA; www.umich.edu) have addressed this limitation by developing a vision-based motion capture technique to detect unsafe actions of individuals captured on videos taken on jobsites.
More specifically, a commercial motion capture system from Vicon (Los Angeles, CA, USA; www.vicon.com) at the University of Michigan's (Ann Arbor, MI, USA; www.umich.edu) 3-D laboratory was first used to capture accurate 3-D motion datasets from individual workers as they performed unsafe actions.
3-D skeleton motion models were also computed from 2-D images taken by surveillance cameras from different viewpoints on the worksite. These 3-D skeleton models were then converted to the same data format as the models created in the laboratory. Next, both models were transformed onto the same space so that the similarity between them could be evaluated.
As a case study, motion data for unsafe actions during ladder-climbing were collected and compared with motion data captured on a site. The result revealed that the vision-based technique worked well, detecting particular unsafe actions of individuals, such as climbing a ladder with a toolkit.
Professors Peña-Mora and Lee believe that such a system could provide valuable feedback to workers, enabling them to assess how safely they are performing a particular job and carry out their tasks in a safer fashion. Since vision-based behavior monitoring does not require any manual effort, it could provide a means to observe workers on a construction site in a cost effective manner.
Vision Systems Articles Archives
