Recognition of High-Risk Scenarios in Building Construction Based on Image Semantics

Abstract

The action analysis and semantic interpretation of images have recently attracted increased attention in the field of computer vision. However, it is difficult for an intelligent monitoring method based on computer vision to understand complex scenarios and describe hazardous events from a surveillance video. To identify risks in a construction process and prevent construction accidents, an automatic identification method combining object detection and ontology is proposed. First, a faster region-convolutional neural network is used to extract low-level semantic information from scene elements and element spatial relationship attributes from images exported from a surveillance video. Second, an ontology semantic network is established within the scope of a construction scene, and logical language of the ontology is used to transform the low-level semantic information of images into high-level semantics of event descriptions. Third, construction risk rules are translated into ontology rules, and high-risk situations that may arise at the construction site are identified by a Pellet inference engine. Finally, a foundation pit excavation scene is taken as an example, and test results are used to verify the feasibility and effectiveness of the proposed method. The proposed method can be used to improve the efficiency of construction safety management.