高速公路拥堵事件检测中的背景建模及状态判别方法研究

发布时间：2018-01-18 02:25

本文关键词：高速公路拥堵事件检测中的背景建模及状态判别方法研究　出处：《重庆大学》2014年硕士论文　论文类型：学位论文

【摘要】：由于场景的封闭性及行车速度高的特点，发生在高速公路的交通拥堵严重影响其通行能力，极易引发二次事故而导致人员伤亡和严重的经济损失。作为基本的监控手段，高速公路关键路段已安装大量的视频监控装置，传统的基于视频的拥堵检测方法由于高速公路场景复杂、视频图像易受环境影响的问题而难以获得满意的结果。因此，充分利用现有监控资源，研究基于视频的高速公路交通拥堵状态判别方法对提高高速公路通行能力和运行安全具有重要的理论和实际意义。车辆目标分割、交通参数的提取是交通拥堵判别的基础，现有基于视频的交通拥堵状态判别对复杂环境下车辆目标分割、目标不完整时交通参数提取以及降低交通拥堵状态判别误检测的问题鲜有研究。针对上述问题，论文在深入分析现有交通拥堵状态判别算法的基础上，重点研究了目标提取和状态判别，包括复杂场景下的背景建模、交通参数提取和降低交通拥堵状态判别误检测的问题，最终形成了一套基于视频的高速公路环境下的交通拥堵状态判别方法。在车辆目标提取方面，针对高速公路场景复杂、视频图像易受环境影响的特点，建立了非参数核密度估计方法的背景建模，同时给出了场景光线突变和渐变条件下的背景更新方案。此外，针对车流密度大而导致获取的背景图像质量较差的问题，提出了基于分形维数的初始车流密度检测算法。在此基础上，利用背景差分法和形态学的前景去噪方法提取车辆目标。实验结果表明，提出的背景建模与更新方法能够判断初始车流密度的大小和适应背景光线的变化，从而提高了提取出的车辆目标的准确性。在交通拥堵状态判别方面，提出了基于模糊C均值的拥堵状态判别方法。针对车辆目标检测不完整而导致交通参数获取不准确的问题，通过对常用交通参数特性的分析，提出了以平均空间占有率和时间占有率作为交通拥堵判别参数。在此基础上，根据交通状态之间具有的模糊性，利用模糊C均值算法来获取拥堵状态的聚类中心，并利用欧式距离来得到当前的交通拥堵状态。此外，为了降低交通拥堵状态的误检测，提出了基于连通域分析的误检测识别算法和基于投票机制的拥堵判别方法，进一步提高了交通拥堵状态判别算法的准确性。最后，，综合上述研究成果，建立了高速公路交通拥堵状态判别实验系统，利用重庆市高速公路监控场景视频，在VC环境下进行实验验证。实验结果表明，论文所提方法能够获得较准确的车辆目标，且仅须提取少量交通参数就能判别交通拥堵状态，提高了高速公路复杂场景下的交通拥堵状态判别的准确性和可靠性。
[Abstract]:Due to the closure of the scene and the characteristics of high speed, traffic congestion on the freeway seriously affects its capacity. It is easy to cause secondary accidents and cause casualties and serious economic losses. As a basic means of monitoring, a large number of video surveillance devices have been installed in key sections of the highway. The traditional video based congestion detection method is difficult to obtain satisfactory results due to the complexity of freeway scene and the vulnerability of video image to environment. Therefore, the existing monitoring resources are fully utilized. It is of great theoretical and practical significance to study the video based identification method for freeway traffic jams to improve highway traffic capacity and operation safety. Vehicle target segmentation and extraction of traffic parameters are the basis of traffic congestion discrimination. The existing video based traffic congestion state discrimination is used to segment vehicle targets in complex environment. The problem of extracting traffic parameters and reducing traffic congestion discrimination error detection when the target is incomplete is rarely studied. In view of the above problems, this paper deeply analyzes the existing traffic congestion discrimination algorithms. The research focuses on object extraction and state discrimination, including background modeling in complex scenarios, traffic parameter extraction and error detection in reducing traffic congestion. Finally, a set of video-based traffic congestion discrimination method is formed. In the aspect of vehicle target extraction, aiming at the complexity of freeway scene and the vulnerability of video image to environment, the background modeling of nonparametric kernel density estimation method is established. At the same time, the background updating scheme under the condition of sudden change of scene light and gradual change is given. In addition, the quality of the background image is poor because of the heavy traffic density. An algorithm for detecting initial vehicle flow density based on fractal dimension is proposed. Based on this, the background difference method and morphological foreground denoising method are used to extract vehicle targets. The proposed background modeling and updating method can judge the initial traffic density and adapt to the change of background light, thus improving the accuracy of the extracted vehicle targets. In the aspect of traffic congestion identification, a method based on fuzzy C-means is proposed, which leads to the inaccurate acquisition of traffic parameters due to the incomplete detection of vehicle targets. Based on the analysis of the characteristics of common traffic parameters, the average space occupation rate and time occupancy rate are used as traffic congestion discrimination parameters, and on this basis, according to the fuzziness between traffic states. The fuzzy C-means algorithm is used to obtain the clustering center of the congestion state, and the Euclidean distance is used to get the current traffic congestion state. In addition, in order to reduce the false detection of the traffic congestion state. An algorithm of error detection and identification based on connected domain analysis and a congestion discrimination method based on voting mechanism are proposed to improve the accuracy of traffic congestion identification algorithm. Finally, based on the above research results, a highway traffic congestion identification experiment system is established. The video of Chongqing expressway monitoring scene is used to verify the experiment in VC environment. The method proposed in this paper can obtain more accurate vehicle targets, and only a small number of traffic parameters can be extracted to identify traffic congestion. The accuracy and reliability of identification of traffic congestion state in complex scene of expressway are improved.
【学位授予单位】：重庆大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U491.265

【参考文献】