基于隐马尔可夫模型的城市道路路段不良驾驶行为鉴别

发布时间：2018-03-13 08:05

  本文选题：城市道路路段　切入点：不良驾驶行为　出处：《哈尔滨工业大学》2016年硕士论文　论文类型：学位论文

【摘要】：根据城市道路历史交通事故统计数据显示,在诱发或造成交通事故的诸多因素中人的因素占主导地位,基于历史事故数据的城市道路安全研究本身属于一种“事后补救型”,而基于非事故数据的“防患未然型”逐步引起交通安全领域的重视,从常见的交通流以及交通冲突有效识别城市道路的不良驾驶行为信息,是评价城市道路不良驾驶安全水平的重要途径之一。本文分析了城市道路路段的交通流基本特性,将交通流状态划分为畅行流、稳定流以及强制流三种状态,与其对应则有自由驾驶行为、跟驰驾驶行为以及换道驾驶行为三种驾驶行为。基于此,对城市道路路段不良驾驶行为进行界定,主要有超速不良驾驶行为、压线不良驾驶行为、违章掉头不良驾驶行为、未保持安全车距不良驾驶行为、频繁换道不良驾驶行为五种。分别对晴天、雨天情境下城市道路路段交通流进行视频数据采集,使用视频处理软件获取车辆的运动轨迹数据,提取每一辆车的运动参数并分析长江路研究区段的交通流特性及驾驶特性,具体分析晴天、雨天不同情境下的交通量及其组成、车辆速度、车辆加速度、车头时距、冲突时距、距右侧车道线距离以及距摄像头距离每一项特性。其中,本文构建的冲突度量指标“冲突时距”是随着速度、加速度、天气等变化而变化的,这弥补了在以往研究中速度、方向或加减速度不变假设条件的缺陷。本文采用隐马尔可夫模型从左至右模型结构,输入观测序列速度、加速度、距右侧车道线距离、冲突时距以及距摄像头距离五项观测序列信息,分别构建晴雨天场景下,正常驾驶行为、超速不良驾驶行为、压线不良驾驶行为、违章掉头不良驾驶行为、未保持安全车距不良驾驶行为以及频繁换道不良驾驶行为六种状态识别模型,并验证模型识别精度高达86%以上。最后,针对本文每一项不良驾驶行为提出了改善惩罚措施,其效果可采用HMM模型进行评价。
[Abstract]:According to the statistics of urban road traffic accidents, human factors play a dominant role in inducing or causing traffic accidents. The study of urban road safety based on historical accident data belongs to a kind of "remedial type" after the event, while the "precautionary type" based on non-accident data gradually attracts the attention of the traffic safety field. It is one of the important ways to evaluate the safety level of bad driving of urban roads from the common traffic flow and the information of traffic conflict to identify the bad driving behavior of urban roads. This paper analyzes the basic characteristics of traffic flow in urban road sections. The traffic flow state is divided into three states: smooth flow, steady flow and forced flow, corresponding to which there are three driving behaviors: free driving behavior, following driving behavior and changing traffic driving behavior. To define the bad driving behavior of urban road sections, there are mainly bad driving behavior in speeding, bad driving behavior in line pressing, bad driving behavior in illegal turn around, bad driving behavior without keeping safe car distance, There are five kinds of bad driving behavior in changing roads frequently. We collect video data of traffic flow on urban road sections in sunny and rainy days, and use video processing software to obtain the moving track data of vehicles. The motion parameters of each vehicle are extracted and the traffic flow characteristics and driving characteristics of the study section of Changjiang Road are analyzed. The traffic volume and its composition, vehicle speed, vehicle acceleration, headway time distance, conflict time distance in sunny and rainy days are analyzed in detail. The distance from the right lane to the right lane and the distance from the camera. Among them, the conflict measure "conflict time distance" constructed in this paper changes with the speed, acceleration, weather and so on, which makes up for the speed in previous studies. In this paper, the structure of hidden Markov model from left to right is used to input the velocity, acceleration and distance from the right driveway of the observation sequence. According to the five observation sequence information of conflict time distance and distance from camera, normal driving behavior, speeding bad driving behavior, line pressing bad driving behavior, illegal turning around bad driving behavior were constructed in rainy and sunny weather respectively. There are six state recognition models of bad driving behavior without keeping safe distance and frequent bad driving behavior of changing lanes, and the recognition accuracy of the model is as high as 86%. Finally, for each bad driving behavior in this paper, some measures are put forward to improve the punishment. Its effect can be evaluated by HMM model.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U492.8
，

本文编号：1605505