基于语义网的挖掘机车斗违规运载工人识别方法研究

周奥

摘  要

  我国每年发生的建筑施工安全事故总体依然呈上升趋势，而导致这些事故发生的主要原因是工人的不安全行为。本文以挖掘机车斗违规运载工人这种不安全行为为例，研究一种新的工人不安全行为视觉识别方法。将计算机视觉应用于识别施工现场工人的不安全行为，能够加强对工人的外部监督，在一定程度上减少安全事故的发生。但目前只依赖视觉的识别方法需要大量含有该行为的训练数据来提取某种特定的行为特征，这不仅存在数据收集和标注的困难，还对视觉算法检测特征的准确度提出了更高的要求。考虑到这一点，本文提出一种结合视觉和语义的识别方法。它无需特定的数据训练，只需准确检测出施工现场的挖掘机车斗和工人，即可利用语义网技术识别出挖掘机车斗违规运载工人。

  该方法的实施包括三个步骤：（1）将建筑工人不安全行为知识表示成语义网络的结构化形式，让计算机能够处理并利用这些知识进行语义识别；（2）基于Faster R-CNN算法从施工图像中检测出挖掘机车斗和工人，以供计算机获取施工现场的视觉检测数据；（3）在语义数据管理系统（图数据库）中集成这两种类型的信息，利用图数据库的数据推理能力识别出挖掘机车斗违规运载工人的不安全行为。这首先要将视觉检测数据中的语义信息提取出来，并使其能够自动建模为待识别的语义网模型；然后将结构化的语义知识按照图数据库的推理语言设计出建筑工人不安全行为的语义识别规则；最后利用图数据库的推理引擎自动查询出待识别的语义网模型中是否存在建筑工人不安全行为的语义数据，进而实现挖掘机车斗违规运载工人的识别。为了验证该方法的有效性，本文随机选取了武汉市某综合体地铁建设项目进行实证分析，验证结果显示从施工现场监控图像中识别出挖掘机车斗违规运载工人的准确率在83.54%~90.72%之间，这较为科学地说明了本文研究的识别方法能够有效地识别出挖掘机车斗违规运载工人的不安全行为。

  本文研究为建筑工人不安全行为的视觉识别方法提供了一种新的研究视角，也为建筑施工现场安全监测工作的信息化提供了新的技术方法。

关键词：安全管理  工人不安全行为  计算机视觉  语义网  图数据库

Abstract

  The annual construction safety accidents in China are still showing an upward trend, and the main cause of these accidents is the unsafe behavior of workers. Taking the unsafe behavior of workers carried by excavator bucket as an example, a new computer vision-based approach to identify worker’s unsafe behavior is studied in this paper. Applying computer vision to identify unsafe behavior of workers on construction site can strengthen the external supervision of workers, which can reduce the occurrence of accidents to a certain extent. However, the existing visual-only recognition methods need a large amount of training data containing the behavior to extract a specific behavioral feature. This not only has difficulty in data collection and labeling, but also requires a higher accuracy for the visual algorithm to detect the features. With this in mind, a new recognition approach integrating sematic and vision is developed in this paper. It can utilize the sematic web technology to identify the unsafe behavior of workers without specific data training, so just detect the objects on the construction site accurately.

   The approach studied in this paper include:

   (1) Representing the knowledge of worker’s unsafe behaviors in a structured form of semantic web, in order to be reused by computer for semantic recognition;

   (2) Detecting excavator buckets and workers from construction image based on Faster R-CNN algorithm, in order that computer can obtain the visual information of construction site;

   (3) Integrating these two types of information in semantic data management system (graph database), and utilizing the data reasoning ability of the graph database to identify the unsafe behavior of worker carried by excavator bucket. First, extracting the semantic data from visual information and modeling it into a semantic web data model automatically. Then, designing the semantic data reasoning rules of worker’s unsafe behaviors with the structural knowledge. Finally, utilizing the data operation ability of the graph database to query the sematic data of the unsafe behavior in the semantic web data model.

    After all these processing, the unsafe behavior of workers carried by excavator bucket can be recognited from the image. In order to verify the validity of this approach, a subway construction project in Wuhan city was randomly selected as a case study in this paper. The results show that the recognition accuracy for the unsafe behavior from the construction site monitoring images is between 83.54% and 90.72%. It showed that the approach studied in this paper can be effectively applied to identify the unsafe behavior of workers carried by excavator bucket on the construction site.

    This paper provides a new perspective for the study of visual identification of worker’s unsafe behaviors, and it also provides a new technical approach for the informationization of safety monitoring on the construction site.

Keywords: Safety management  Unsafe behavior of worker  Computer vision  Sematic web  Graph database