Multi-Angle Fusion-Based Safety Status Analysis of Construction Workers

Hazardous accidents often happen in construction sites and bring fatal consequences, and therefore safety management has been a certain dilemma to construction managers for long time. Although computer vision technology has been used on construction sites to identify construction workers and track their movement trajectories for safety management, the detection effect is often influenced by limited coverage of single cameras and occlusion. A multi-angle fusion method applying SURF feature algorithm is proposed to coalesce the information processed by improved GMM (Gaussian Mixed Model) and HOG + SVM (Histogram of Oriented Gradient and Support Vector Machines), identifying the obscured workers and achieving a better detection effect with larger coverage. Workers are tracked in real-time, with their movement trajectory estimated by utilizing Kalman filters and safety status analyzed to offer a prior warning signal. Experimental studies are conducted for validation of the proposed framework for workers’ detection and trajectories estimation, whose result indicates that the framework is able to detect workers and predict their movement trajectories for safety forewarning.