Motion Recognition Method for Construction Workers Using Selective Depth Inspection and Optimal Inertial Measurement Unit Sensors

Abstract

The construction industry holds the worst safety record compared to other industrial sectors, and approximately 88% of accidents result in worker injury. Meanwhile, after the development and wide application of deep learning in recent years, image processing has greatly improved the accuracy of human motion detection. However, owing to equipment limitations, it is difficult to effectively improve depth-related problems. Wearable devices have also become popular recently, but because construction workers generally work outdoors, the variable environment makes the application of wearable devices more difficult. Therefore, reducing the burden on workers while stabilizing the detection accuracy is also an issue that needs to be considered. In this paper, an integrated sensor fusion method is proposed for the hazard prevention of construction workers. First, a new approach, called selective depth inspection (SDI), was proposed. This approach adds preprocessing and imaging assistance to the ordinary depth map optimization, thereby significantly improving the calculation efficiency and accuracy. Second, a multi-sensor-based motion recognition system for construction sites was proposed, which combines different kinds of signals to analyze and correct the movement of workers on the site, to improve the detection accuracy and efficiency of the specific body motions at construction sites.