高速公路中车辆对象同一性的检索
发布时间：2018-07-17 08:54
【摘要】：智能交通(Intelligent Transportation System)是未来交通系统发展的方向,它将先进的交通控制理论和计算机科学理论有效应用到交通管理系统中,服务于人类。高速公路中的未悬挂牌照或车牌遮挡等无法识别车辆(本文以下统称灰牌照目标车辆,简称目标车辆)的问题是当前智能交通中亟待解决的热点为题。 针对该问题,本论文围绕图像和视频中目标车辆对象的同一性比较和目标车辆对象的时空关联检索展开：利用目标车辆对象在图像、视频中的颜色、车类、局部不变特征等信息进行同一性比对；利用目标车辆所在的卡口地理位置、时间、速度、运动方向等特征同时结合同一性检测方法进行目标车辆对象的时空关键检索,确定概率条件下可信的目标车辆相似序列。本论文的具体研究内容如下： 1.从车辆对象检测与特征提取、车辆对象同一性比较和视频检索三个方面介绍国内外研究现状。结合现有理论成果,论述目标车辆进行时空关键检索研究思路与现实意义; 2.车辆对象的检测和特征提取。实现高速公路监控平台中视频、图像的目标车辆对象的检测和车脸区域的提取；并提取车辆对象的车类、颜色特征,运动方向特征、局部不变特征及PHash等特征； 3.融合多种特征的车辆对象同一性比较算法研究。对目标车辆所在的视频、与图像进行ROI区域的分辨率实现尺度归一化,对目标车辆的车类、颜色、局部及Phash区域特征实现特征归一化,利用分层判别法筛选出与目标车对象相似的车辆对象； 4.时空关联检索。结合目标车辆的时空信息,建立目标车辆对象的时空关联检索模型,从正向、反向时空关联分析视频中的车辆对象的相似性,最终达到目标车辆对象在高速公路中多个相关卡口的同一性检索。 在以上研究的基础上,构建一个针对“灰牌照”目标车辆车型识别与视频关联检索的研究原型,可以对目标车辆在高速公路上进行时空关联的视频分析,建立该目标车辆在途卡口的视频关联检索的理论基础,为公安交警对“灰牌照”违法车辆的追溯和违法证据取证提供参考。
[Abstract]:Intelligent Transportation system (Intelligent Transportation system) is the development direction of traffic system in the future. It effectively applies advanced traffic control theory and computer science theory to traffic management system and serves human beings. The problem that the vehicle can not be recognized such as the unsuspended license plate or the license plate occlusion in the expressway (hereinafter referred to as the target vehicle with grey license plate) is a hot topic to be solved in the intelligent traffic. To solve this problem, this paper focuses on the identity comparison of the target vehicle object in image and video and the spatio-temporal association retrieval of the target vehicle object: using the color of the target vehicle object in the image, the video, the vehicle class, the target vehicle object in the image, the video color, the vehicle class. Using the location, time, speed, direction of movement and other features of the bayonet where the target vehicle is located, and combining with the identity detection method, the spatio-temporal key retrieval of the target vehicle object is carried out. A credible target vehicle similarity sequence is determined under probabilistic conditions. The specific contents of this thesis are as follows: 1. This paper introduces the research status of vehicle object detection and feature extraction, vehicle object identity comparison and video retrieval. Combined with the existing theoretical results, this paper discusses the research ideas and practical significance of time-space key retrieval for target vehicles. 2. Vehicle object detection and feature extraction. The vehicle object detection and vehicle face region extraction of video and image in highway monitoring platform, the vehicle class, color feature, motion direction feature, local invariant feature and PHash feature of vehicle object are extracted. 3. Research on vehicle object identity comparison algorithm with multiple features. The resolution of ROI region is normalized for the video and image of the target vehicle, and the vehicle class, color, local and Phash region features of the target vehicle are normalized. The hierarchical discriminant method is used to screen the vehicle object similar to the target vehicle object. 4. Time and space connection search. Combining the spatio-temporal information of the target vehicle, the spatio-temporal association retrieval model of the target vehicle object is established, and the similarity of the vehicle object in the video is analyzed from the forward and backward time-space connection. Finally, the identity retrieval of the target vehicle object in multiple related bayonets in the expressway is achieved. On the basis of the above research, a research prototype of "gray license plate" vehicle model recognition and video association retrieval is constructed, which can be used to analyze the video of the target vehicle on the freeway. The theoretical basis of video correlation retrieval of the target vehicle in the road bayonet is established, which provides a reference for the police to trace the illegal vehicle with "grey license plate" and to obtain evidence of the violation.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U495
[Abstract]:Intelligent Transportation system (Intelligent Transportation system) is the development direction of traffic system in the future. It effectively applies advanced traffic control theory and computer science theory to traffic management system and serves human beings. The problem that the vehicle can not be recognized such as the unsuspended license plate or the license plate occlusion in the expressway (hereinafter referred to as the target vehicle with grey license plate) is a hot topic to be solved in the intelligent traffic. To solve this problem, this paper focuses on the identity comparison of the target vehicle object in image and video and the spatio-temporal association retrieval of the target vehicle object: using the color of the target vehicle object in the image, the video, the vehicle class, the target vehicle object in the image, the video color, the vehicle class. Using the location, time, speed, direction of movement and other features of the bayonet where the target vehicle is located, and combining with the identity detection method, the spatio-temporal key retrieval of the target vehicle object is carried out. A credible target vehicle similarity sequence is determined under probabilistic conditions. The specific contents of this thesis are as follows: 1. This paper introduces the research status of vehicle object detection and feature extraction, vehicle object identity comparison and video retrieval. Combined with the existing theoretical results, this paper discusses the research ideas and practical significance of time-space key retrieval for target vehicles. 2. Vehicle object detection and feature extraction. The vehicle object detection and vehicle face region extraction of video and image in highway monitoring platform, the vehicle class, color feature, motion direction feature, local invariant feature and PHash feature of vehicle object are extracted. 3. Research on vehicle object identity comparison algorithm with multiple features. The resolution of ROI region is normalized for the video and image of the target vehicle, and the vehicle class, color, local and Phash region features of the target vehicle are normalized. The hierarchical discriminant method is used to screen the vehicle object similar to the target vehicle object. 4. Time and space connection search. Combining the spatio-temporal information of the target vehicle, the spatio-temporal association retrieval model of the target vehicle object is established, and the similarity of the vehicle object in the video is analyzed from the forward and backward time-space connection. Finally, the identity retrieval of the target vehicle object in multiple related bayonets in the expressway is achieved. On the basis of the above research, a research prototype of "gray license plate" vehicle model recognition and video association retrieval is constructed, which can be used to analyze the video of the target vehicle on the freeway. The theoretical basis of video correlation retrieval of the target vehicle in the road bayonet is established, which provides a reference for the police to trace the illegal vehicle with "grey license plate" and to obtain evidence of the violation.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U495
【参考文献】
相关期刊论文 前10条
1 敖成龙,苏英,龚元明;基于相似度的复杂数据对象比较[J];北京理工大学学报;2003年05期
2 徐蕾;潘晓璐;李一民;赵晓辉;;基于特征点光流的车辆检测[J];重庆理工大学学报(自然科学);2011年09期
3 沈\，
本文编号：2129892
本文链接：https://www.wllwen.com/kejilunwen/jiaotonggongchenglunwen/2129892.html
