In order to ensure the safety of construction, all kinds of construction machinery are widely applied to the construction site. Tower crane, as a material handling equipment, has the characteristics of wide operating range and large potential energy, and has become the core machinery in the construction site. The tower crane driver’s field of vision is often blocked, which seriously affects the safety of hoisting. To increase the view of tower crane drivers, most of the current monitoring systems will install a camera on the boom above the hook. But this camera can only view the situation around the hook, and it cannot be quantified. Based on this, this paper proposes a hoisting security detection technology based on deep learning. Firstly, the camera in the monitoring system is used to collect data sets. Secondly, the hook and workers are marked in the image. Then, Faster R-CNN is used to train and evaluate the data sets. The results show that the method has high recognition accuracy. However, the worker and the hook are not on a horizontal plane, so a verification test of the relationship between the height and the ratio of pixel length to true length was completed. The results show that the method can convert the ratio of the hook to the ratio of the worker, and then the real distance between the worker and the hook can be calculated.
As an effective fall accident preventive method, insight into near-miss falls provides an efficient solution to find out the causes of fall accidents, classify the type of near-miss falls and control the potential hazards. In this context, the paper proposes a method to detect and identify near-miss falls that occur when a worker walks in a workplace based on artificial neural network (ANN). The energy variation generated by workers who meet with near-miss falls is measured by sensors embedded in smart phone. Two experiments were designed to train the algorithm to identify various types of near-miss falls and test the recognition accuracy, respectively. At last, a test was conducted by workers wearing smart phones as they walked around a simulated construction workplace. The motion data was collected, processed and inputted to the trained ANN to detect and identify near-miss falls. Thresholds were obtained to measure the relationship between near-miss falls and fall accidents in a quantitate way. This approach, which integrates smart phone and ANN, will help detect near-miss fall events, identify hazardous elements and vulnerable workers, providing opportunities to eliminate dangerous conditions in a construction site or to alert possible victims that need to change their behavior before the occurrence of a fall accident.