基于视频分析的道路车辆信息提取方法研究

发布时间：2018-03-08 23:17

本文选题：视频分析　切入点：车辆信息　出处：《海南大学》2017年硕士论文　论文类型：学位论文

【摘要】：道路上的车辆是智能交通管理系统中的重点研究对象,基于视频分析的车辆信息的获取,能够有效的协助交通部门监管,为人民大众提供更加高效、安全的出行条件。目前,虽然在道路车辆监测方面已有大量的研究成果,但大多数基于视频图像的车辆监控系统功能单一、对摄像头的要求不一致,并且部分信息提取时准确率不够高。本文针对上述问题,主要研究内容如下:(1)针对目前车辆监控系统功能单一、对摄像头的要求不一致的问题,本文利用目前国内道路上较多的一类摄像头采集视频,分析在该摄像头下的十字路口场景,根据场景整体的设计了提取车辆的个体特征和群体行为特征的系统框架,使得车流量统计、车速测算、车标识别、排队拥堵检测这些功能之间能够互相协同、节约资源,并对功能实现时需要用到的图像处理和机器学习的基本方法进行了比较分析。(2)设计了车辆定位和跟踪的算法,并以此为基础实现了车流量统计和车速测算。本文改进Adaboost联合SVM的方法,实现对车牌精确实时的定位;选取时空上下文视觉跟踪算法跟踪车牌,解决处理速度慢、不准确的问题;依据V=S/T,分段计算轨迹并乘以不同比例系数将轨迹长度转换为实际路程,实现车辆测速。(3)提出多尺度并行卷积神经网络对车标进行分类,在较少训练数据集的情况下达到较高识别准确率,并与其它方法进行对比,验证其实用性,·优化现有的车标定位算法,使之适应本系统所要应用的场景。(4)结合背景差法和SVM分类器来实现排队拥堵检测。首先分析场景,设计虚拟检测窗口;然后判断每个检测窗口是否有车辆和前景;最后结合前景所占比率和是否有车辆来判断窗口是否有停车,从而判断出整体排队长度和溢出情况。本文在同一摄像头采集到的同一视角的场景下,根据实际条件和需求实现了对车辆个体特征和群体行为特征提取并将结果以语义的方式输出,文章对各个功能的算法进行了实际测试,验证了其准确性和鲁棒性。本文实现的系统可在原视频中基于标注点快速进行定位,将视频的高层语义信息以图文方式展现给用户。
[Abstract]:The vehicle on the road is a key research object of intelligent traffic management system, access to vehicle information based on video analysis, can effectively assist the traffic supervision department, for the people to provide more efficient and safe travel conditions. At present, although there are a lot of research achievements in road vehicle monitoring, but most video vehicle monitoring system the image features based on a single camera, the requirements are inconsistent, and part of the information extraction accuracy is not high enough. Aiming at the above problems, the main research contents are as follows: (1) aiming at the vehicle monitoring system of single function, the camera requirements is not the same problem, the current domestic road more of a class of camera video capture, camera analysis in the crossroads of the scene, according to the characteristics of individual and group behavior characteristics of the overall design of the scene extraction vehicle The framework of the system, the traffic flow statistics, speed calculation, vehicle logo recognition, mutual collaboration, between the queue congestion detection of these features of saving resources, basic methods and Study on image processing and machine functions needed are compared and analyzed. (2) the design of the vehicle positioning and tracking algorithm, and then based on the realization of the traffic flow statistics and speed calculating. This improved method of Adaboost combined with SVM, to realize the accurate and real-time license plate location; select the spatio-temporal context visual tracking algorithm to track the license plate, to solve the slow processing speed, the problem of inaccurate; according to V=S/T, calculate the trajectory and multiplied by the different proportion coefficient will be converted to the actual distance of piecewise trajectory length and realize vehicle speed. (3) we propose a multi-scale parallel convolutional neural network to classify the logo, to achieve higher recognition accuracy with less training data set, and Compared to other methods, to verify its practicality, and optimize the existing logo positioning algorithm, so as to adapt to the system by the application of the scene. (4) combined with background difference method and SVM classifier to achieve queue congestion detection. Firstly, design of virtual scene, and then judge the detection window; each detection window is a vehicle and finally, future prospects; the proportion and whether there is a vehicle to determine whether the window is parking, in order to determine the overall queue length and overflow. The same view in the same camera to capture the scene, according to the actual conditions and needs of the vehicle to achieve the individual characteristics and group behavior and will result in feature extraction the semantics of the output, the algorithm of each function was tested to verify its accuracy and robustness. This system can be implemented in the original video annotation based on point quickly Positioning, the high-level semantic information of the video is presented to the user in a graphic manner.

【学位授予单位】：海南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U495;TP391.41

【相似文献】