基于本体和描述逻辑的交通事件语义表现方法研究

发布时间：2018-12-16 17:05

【摘要】：基于本体和描述逻辑的交通事件语义识别和表达是智能交通领域的研究热点,能为城市交通管理和控制提供有效、完整、准确的实时交通状态信息,对于减轻交通压力,减少交通事故,提高交通运行效率和城市交通智能化水平均具有指导意义。然而,现有的计算机图像语义分析技术无法自动识别交通视频图像序列表达的交通事件高级语义,主要原因:一是缺乏对象的概念层次识别方法;二是低级的图像视觉概念映射到事件高级语义时存在“语义鸿沟”问题。这使得传统的交通场景事件理解方法大多忽略了语义分析,仅针对纯粹的图像数据进行分析。因此,道路交通事件内容的高级语义识别、表达和推理方法研究,是一项富有重要研究意义的课题。本文源于省基金资助项目“基于交通态势评估的道路安全推理研究”子课题的研究成果,共分三个部分:概念层次的对象识别方法研究、低级概念映射到高级语义的方法研究以及交通事件高级语义的识别、表达和推理系统开发。本文主要贡献如下:(1)利用改进的“LST-KDE”算法、基于HSI颜色空间的Camshift算法以及Hough变换等提取交通图像中对象的特征属性(颜色、纹理等);然后,借鉴领域本体理论,构建交通图像层次概念模型,将交通场景图像中拥有一定属性特征的对象映射为基本概念;最后,给出交通领域本体自动构建方法,建立了交通领域本体知识库,实现了交通场景事件中对象的概念层次识别,并为下文语义推理提供了概念基础。(2)为了解决“语义鸿沟”问题,以描述逻辑作为逻辑表达和推理的基础,利用基于谓词逻辑和Tableau算法的描述逻辑表达式搭建低级概念与高级语义之间的桥梁,提出了利用低级概念描述高级语义的表达方法。其中,描述逻辑角色集是概念映射到语义以及语义推理的关键。为此,引入RCC8空间拓扑和锥形空间方向模型,描述场景对象之间的空间拓扑和方向关系,以此作为描述逻辑的角色集的主要组成部分。(3)提出了交通事件语义识别、推理和表达系统,包括属性融合、语义映射及事件识别和表达3个模块。首先采用训练好的最佳样本图像序列将完整交通事件划分为若干子事件,构建子事件的样本图像序列库;然后利用FSM自动机以及描述逻辑RQL查询语言自动推理并识别每个子事件,最终实现完整交通事件的语义识别和表达。本研究的特点及创新性:使用本体和描述逻辑等能够被计算机识别和处理的形式化自然语言,构建了结构化的语义表达系统;通过模拟人脑思维进行交通场景图像的语义分析,识别并推理场景中交通事件态势演化全过程,为交通场景理解、交通图像检索以及交通事件语义识别和表达提供新的思路和方法。
[Abstract]:Semantic identification and representation of traffic events based on ontology and description logic is a research hotspot in the field of intelligent transportation. It can provide effective, complete and accurate real-time traffic state information for urban traffic management and control, and reduce traffic pressure. It is of guiding significance to reduce traffic accidents, improve traffic efficiency and intelligent level of urban traffic. However, the existing computer image semantic analysis technology can not automatically identify the high-level semantics of traffic events expressed by traffic video image sequence. The main reasons are as follows: first, the lack of object conceptual hierarchy recognition method; Second, there is a semantic gap problem when the low-level image vision concept is mapped to the event high-level semantics. This makes the traditional traffic scene event understanding methods mostly ignore the semantic analysis and only analyze the pure image data. Therefore, the study of high-level semantic recognition, representation and reasoning of road traffic events is an important research topic. This paper derives from the research results of the project "Road Safety reasoning Research based on Traffic situation Assessment", which is funded by the provincial fund. It is divided into three parts: the research of object recognition method at conceptual level, The method of mapping low-level concepts to high-level semantics, and the identification, representation and reasoning system development of traffic event high-level semantics. The main contributions of this paper are as follows: (1) using the improved "LST-KDE" algorithm, the Camshift algorithm based on HSI color space and Hough transform to extract the feature attributes (color, texture, etc.); Then, using domain ontology theory as reference, the hierarchical conceptual model of traffic images is constructed, and the objects with certain attributes in traffic scene images are mapped to basic concepts. Finally, the automatic construction method of traffic domain ontology is given, the knowledge base of traffic domain ontology is established, and the concept hierarchy recognition of objects in traffic scene events is realized. It provides the conceptual basis for the following semantic reasoning. (2) in order to solve the problem of "semantic divide", descriptive logic is used as the basis of logic expression and reasoning. The description logic expression based on predicate logic and Tableau algorithm is used to build the bridge between the low-level concept and the high-level semantics, and a representation method is proposed to describe the high-level semantics by using the low-level concept. Among them, descriptive logical role set is the key to mapping concept to semantic and semantic reasoning. In this paper, RCC8 spatial topology and conical spatial orientation model are introduced to describe the spatial topology and directional relationship between scene objects, which is regarded as the main component of the role set of description logic. (3) Traffic event semantic recognition is proposed. The reasoning and representation system includes three modules: attribute fusion, semantic mapping, event recognition and representation. Firstly, the complete traffic events are divided into several sub-events by using the best sample image sequence, and the sample image sequence database of the sub-event is constructed. Then, FSM automata and description logic RQL query language are used to automatically infer and recognize each sub-event, and finally complete traffic event semantic identification and representation are realized. The characteristics and innovation of this study are as follows: a structured semantic representation system is constructed using formal natural languages such as ontology and description logic which can be recognized and processed by computer. The semantic analysis of traffic scene image is carried out by simulating human brain thinking, and the whole process of traffic event situation evolution in the scene is recognized and inferred, so that the traffic scene can be understood. Traffic image retrieval and semantic identification and representation of traffic events provide new ideas and methods.
【学位授予单位】：山东理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495

【相似文献】