基于本体和视觉的基坑内支撑上违规行走

风险识别研究

赵能

摘要

基坑工程是高层建筑、大型市政等工程施工的重要环节，其中深基坑开挖过程的内支撑系统围护方式已经得到广泛应用。然而，在施工过程中，工人常常将平衡土压力的基坑内支撑结构作为一种通道或临时作业场所，这无疑是一种极易引起高处坠落事故的不安全行为。控制这种不安全行为的传统手段依赖人为监管和管理程序的改善，在实际应用中效率较低，不能及时给工人提供反馈，因此需要提出新的研究方法以实现基坑内支撑上违规行走风险的自动识别。

本文针对基坑内支撑上违规行走场景，结合计算机视觉在自动提取图像语义信息以及本体在知识表达、推理方面的优势，提出了一种风险自动识别方法。首先，利用基于Mask R-CNN（Mask Region-Based Convolutional Neural Network）的施工场景目标检测模型提取图像中的目标属性信息。其次，通过对风险识别过程和不安全场景的分析建立了风险本体知识库，实现了施工安全知识的共享和语义表达，并将目标检测模型提取的信息转化成知识库中的本体实例。最后，对应相关施工安全规范建立了SWRL（Semantic Web Rule Language）规则，将本体知识和规范一起输入Drools推理引擎，对实例存在的风险进行识别判断。

本文开发的施工场景目标检测模型对图像语义信息的提取有较好的准确性，同时通过实例验证了该风险识别方法的可行性。本研究有助于扩展以知识为中心的风险自动识别方法，可以提升施工安全管理水平，并在一定程度上解决了图像分析中底层特征和高级语义之间的“语义鸿沟”问题。

关键词：风险识别；不安全行为；基坑内支撑；违规行走；计算机视觉；本体

Abstract

Deep foundation engineering is an important component in the construction of high-rise buildings and large-scale municipal engineering, and widely use the interior supporting structure to balance the earth pressure. However, site personnel often take it as a kind of passageway or temporary workplace in the construction process. Traditional security supervision methods rely on manual monitoring which are inefficient. Therefore, we need to develop new methods to realize the automatic identification of the risk.

Combined with the advantages of computer vision in automatic extraction of image semantic information and ontology in knowledge expression and reasoning, this paper has proposed a method of automated identification of walking on supports. First of all, We have utilized an object detection model based on the Mask R-CNN (Mask Region-based Convolutional Neural Network) to extract the target attribute information in the images. Secondly, risk ontology knowledge base has been established by analyzing the risk identification process and unsafe scenario to realize the sharing and semantic expression of construction safety knowledge. And the information extracted from the object detection model is transformed into an ontology instance in the knowledge base. Finally, SWRL (Semantic Web Rule Language) rules have been established in accordance with relevant construction safety specifications, and ontology knowledge and specifications have been input into Drools inference engine to identify and judge the risks existing in instances.

The object detection model developed in this paper has a good accuracy in extracting image informations, and the feasibility of the risk identification method has been verified by an example, which is helpful to expand the knowledge-centered risk identification method and improve the construction safety management level. And it can partly solve the “semantic gap” between low-level features extracted from the image and high-level semantic meaning that people can recognize on the image.

Keywords: Risk Identification; Unsafe Behavior; Strut of Construction Pit; Unsafe Walking: Computer Vision; Ontology