基于自动车牌识别数据的城市道路行程时间估计

发布时间：2018-06-24 09:55

本文选题：自动车牌识别数据 + 路段行程时间　；参考：《浙江大学》2017年博士论文

【摘要】：作为描述道路交通状态的一项重要参数,行程时间的相关研究一直以来都是交通工程与交通科学领域的研究热点。准确、实时且可靠的行程时间信息,包括路段行程时间和路径行程时间,是先进的交通管理系统和先进的出行者信息系统的重要支撑。然而交通需求的波动性(如受季节效应、人口特征、交通信息等的影响)、交通供给的波动性(如受交通事件、道路施工、天气因素、道路集合条件等的影响)以及信号控制交叉口车辆到达和离开的随机性,导致城市道路路段行程时间在时间上、空间上以及不同方向上都具有较大的不确定性。因此,对于传统的基于交通流模型的路段行程时间估计方法,其行程时间估计结果误差较大,难以反应实际道路交通状态。随着交通信息采集技术和处理技术的快速发展,交通数据在传统的环形感应线圈数据、微波雷达数据、红外数据的基础上,出现了浮动车数据、自动车辆识别数据、自动车牌识别数据和蓝牙数据等新型交通数据。其中基于高清智能卡口的自动车牌识别数据包含有通过车辆的车牌号、通过时刻、进口方向和进口道编号等数据,继而可以获得流量、单车行程时间及单车行驶方向等信息,而且高清智能卡口系统的布设日益广泛。因此,本文利用基于高清智能卡口的自动车牌识别数据对城市道路路段行程时间以及路径行程时间进行研究。首先对基于高清智能卡口的自动车牌识别数据进行数据质量分析:介绍了高清卡口智能系统的工作原理、布设位置、检测数据以及系统性能指标,在此基础上展开了断面数据质量分析(包括流量精度和自动车牌识别精度)和路段行程时间数据质量分析。将封闭路段作为研究对象,通过分析路段行程时间估计结果(包括路段行程时间估计值和标准差的波动性)与路段行程时间样本率的变化关系,发现样本率越大,路段行程时间估计值的平均绝对百分误差越小,标准差的波动性越小;当样本率大于0.414时,基于样本数据的路段行程时间参数满足精度及稳定性要求,继而确定了路段行程时间的样本率阈值。将非封闭路段作为研究对象,考虑路段开口,计算路段行程时间的实际匹配率,对其时空变化特征和显著性差异进行分析。实际数据表明天气良好时,行程时间的实际匹配率与观测路段及观测日期无关,该值稳定且均大于最小样本率0.414;行程时间的实际匹配率与观测时段有关,20:00～6:00时段内行程时间的实际匹配率相对于一天内其它时段低,但仍大于最小样本率;最终确定了天气良好时高清智能卡口数据用于估计城市道路路段行程时间的可行性。其次,考虑交通流的不同方向,对路段行程时间进行估计:根据交通流在上游交叉口的驶入方向和在下游交叉口的驶离方向,将路段交通流分为9种。受交通需求/供给的波动性以及信号控制交叉口车辆到达和离开的随机性等原因影响,同一路段上不同方向交通流的路段行程时间可能会有所不同,利用实际采集的自动车牌识别数据,对同一路段上不同方向交通流的路段行程时间进行了一系列对比分析,验证了显著性差异的存在;并融合行程时间回归模型,提出了基于交通流方向的路段行程时间估计方法,实现了部分交通流数据缺失时的行程时间估计;通过实际数据分析,验证了估计方法能够有效地处理噪声数据,并且在数据缺失时,估计结果能够较为准确地反映实际交通状态。最后,基于路径行程时间信息的分类和融合,提出了路径行程时间分布的估计方法:根据路段交通流的定义对路径进行重新定义,并进行观测数据提取,实现部分无代表性数据的剔除;利用车辆行驶方向等信息对其路线进行判别,而路线缺口较大的车辆,对其路线进行拆分而非直接判别;在路线判别的基础上,对部分路径行程时间进行扩大,同样忽略路径缺口较大的车辆;根据行程时间的计算方式以及是否完整,将所有路径下行程时间分为两大类,完整的路径行程时间(TTC)和部分路径行程时间(TTP);不同类别的行程时间处理方法不同,当TTC比例较高时,将其经验分布(TTCD)作为路径行程时间分布的估计结果,当TTC比例较低时,利用霍普金斯统计量寻找实验路径上的断点交叉口,将各断点之间部分路径的行程时间分布的卷积作为基于TTP数据的行程时间分布(TTPD),并将TTC数据与基于TTP数据的行程时间分布TTPD进行融合,得到路径行程时间分布的估计结果;并利用实际路网和仿真环境下的256个实例,进行了不同算法路径行程时间估计结果误差分析、识别精度的影响分析、参数mr的影响分析、路径属性的影响分析以及路径行程时间估计结果分析,对本文路径行程时间分布估计模型进行了全面的评价,验证了估计方法较其它方法的有效性。
[Abstract]:As an important parameter for describing road traffic state, the study of travel time has always been a hot spot in the field of traffic engineering and traffic science. Accurate, real-time and reliable travel time information, including road travel time and path travel time, is an advanced traffic management system and advanced traveler information system. Important support. However, the volatility of traffic demand (such as seasonal effects, demographic characteristics, traffic information, etc.), the volatility of traffic supply (such as the impact of traffic events, road construction, weather factors, road assembly conditions, etc.) and the randomness of the arrival and departure of the vehicles at the signal control intersection, resulting in the travel of the urban road section. Time is more uncertain in time, space and different directions. Therefore, for the traditional road travel time estimation method based on traffic flow model, the error of the travel time estimation result is large and it is difficult to respond to the actual road traffic state. With the rapid development of traffic information collection technology and processing technology, the traffic information acquisition technology and processing technology are developed. On the basis of the traditional loop induction coil data, microwave radar data, and infrared data, the datum appeared the floating car data, automatic vehicle identification data, automatic license plate recognition data and Bluetooth data. The automatic vehicle identification data based on high definition intelligent card port include the license plate number passing through the vehicle, through the vehicle license number, At the moment, the import direction and the import number and so on, then we can get the information of the traffic, the travel time and the direction of the single car. Moreover, the high definition intelligent card port system is widely distributed. Therefore, this paper uses the automatic license plate recognition data based on the high-definition intelligent card port for the travel time and the path travel time of the urban road section. Firstly, the data quality analysis of automatic license plate recognition data based on high definition intelligent card port is carried out. The working principle, layout position, detection data and system performance index of high definition card mouth intelligent system are introduced. On this basis, the quality analysis of cross section data (including flow accuracy and automatic license plate recognition accuracy) is developed. The link travel time data quality analysis. The closed section is taken as the research object. By analyzing the relationship between the estimate of the travel time of the section (including the estimated value of the travel time and the fluctuation of the standard deviation) and the sample rate of the section travel time, it is found that the larger the sample rate is, the more the average absolute percentage error of the estimated value of the road travel time is. When the sample rate is greater than 0.414, when the sample rate is greater than 0.414, the link travel time parameters based on the sample data meet the requirements of precision and stability. Then the sample rate threshold of the section travel time is determined. The non closed section is taken as the research object, and the actual matching rate of the section travel time is calculated and the time and space of the section travel time is calculated. The actual data show that the actual matching rate of the travel time is independent of the observed section and the observation date when the weather is good, and the value is stable and greater than the minimum sample rate of 0.414; the actual matching rate of the travel time is related to the observation period, and the actual matching rate of the travel time within the 20:00 to 6:00 period is relative. At the rest of the day, it is low, but still larger than the minimum sample rate. Finally, the feasibility of the high definition intelligent card data is used to estimate the travel time of the urban road section. Secondly, considering the different directions of the traffic flow, the travel time of the section is estimated: according to the direction of the traffic flow in the upstream intersection and the downstream. The direction of departure of the intersection is divided into 9 kinds of traffic flow, which are influenced by the fluctuation of traffic demand / supply and the randomness of the arrival and departure of the vehicle at the intersection. The travel time of the traffic flow in different directions on the same section may be different. A series of contrasts and analyses on the travel time of traffic flow in different directions on the road have been carried out to verify the existence of significant difference, and the estimation method of road travel time based on the direction of traffic flow is put forward by combining the travel time regression model, and the estimation of the travel time of some traffic flow data is realized, and the analysis of the time of the travel time is realized by the actual data analysis. It is proved that the estimation method can effectively deal with the noise data, and the estimation results can reflect the actual traffic state more accurately when the data is missing. Finally, based on the classification and fusion of path travel time information, the estimation method of path travel time distribution is proposed. The path is redefined according to the definition of the road traffic flow. It also carries out the extraction of the observation data, realizes the elimination of some non representative data, uses the vehicle direction and other information to distinguish its route, and the vehicle with the larger path gap is divided instead of directly judging the route. On the basis of the route discrimination, the route travel time is enlarged and the path gap is ignored. Large vehicles; according to the way of calculation and completeness of travel time, the travel time of all paths is divided into two categories, complete path travel time (TTC) and partial path travel time (TTP); different types of travel time processing methods are different, and when the proportion of TTC is high, their experience distribution (TTCD) is used as path travel time distribution. When the ratio of TTC is low, the Hopki statistic is used to find the intersection of the breakpoint on the experimental path, and the convolution of the travel time distribution of the partial path between the breakpoints is used as the travel time distribution (TTPD) based on the TTP data, and the TTC data is fused with the travel time distribution based on the TTP data, and the path is obtained. The estimated result of the travel time distribution, and using 256 examples under the actual road network and the simulation environment, the error analysis of the path travel time estimation results of different algorithms, the influence analysis of the recognition accuracy, the influence analysis of the parameter Mr, the influence analysis of the path attribute and the analysis of the path travel time estimation results are made. A comprehensive evaluation of the inter estimation model is carried out to verify the effectiveness of the estimation method compared with other methods.
【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：U491

【参考文献】