【作者】 王川童；

【导师】 孙棣华；

【作者基本信息】 重庆大学 ， 控制理论与控制工程， 2010， 硕士

【摘要】 交通拥堵判别技术是及时掌握城市道路交通拥堵状况的重要方式之一。视频检测技术由于具有处理信息量大、无需破坏地面、安装维护方便等优点,因此,在道路交通拥堵判别方面具有较好的应用前景。现有交通视频检测技术一般侧重于道路交通事件检测或交通参数提取,已取得了良好的进展,但尚未用于城市道路交通拥堵的实时判别。因此,充分发挥视频检测技术的优势,研究从整体上实时地判别城市道路交通拥堵状况的方法,对于提高交通视频检测技术的实用性具有重要意义。论文在深入分析基于视频处理的城市道路交通拥堵判别的问题和难点后,重点研究基于视频的交通特征参数获取和交通拥堵判别算法两部分内容。在基于视频的交通特征参数获取方面,针对城市道路的复杂情况,选取了基于非参数核密度的背景建模方法并提出了基于概率比较的目标去噪方法实现车辆目标的提取,然后通过卡尔曼滤波跟踪与虚拟检测线法获取交通特征参数。与传统的背景建模方法相比,基于非参数核密度的背景建模方法具有更好的建模效果,更适应复杂场景的建模。提出的基于概率比较的目标去噪方法能有效去除路边抖动树叶造成的干扰,较好地实现了车辆目标的检测提取,进而提高了车辆识别的精度。在交通拥堵判别算法方面,改进了传统的基于模糊综合判别的交通拥堵判别模型。本文在分析交通特征参数变化特征的基础上,提出先根据速度进行道路交通拥堵预判别,建立不同预判别等级下的参数权重值集合,然后结合预判别结果、实际判别结果和历史数据获得最终的交通拥堵判别结果。由于本文考虑了交通特征参数在不同的交通状态下具有不同的重要性,因此,本文方法能更好地拟合城市道路交通拥堵判别情况。最后,结合提出的基于视频的交通特征参数提取和交通拥堵判别算法,建立了基于视频处理的城市道路交通拥堵判别实验系统,并利用重庆市的城市道路视频监控数据,在VC环境下进行了验证实验。结果表明,给出的方法能较准确地提取交通特征参数,改进的拥堵判别模型比直接用模糊综合判别算法二级跳变率更低,能实时、高效地实现城市道路交通拥堵判别,具有较好的准确性及可行性。更多还原

【Abstract】 Traffic congestion identification technology is one of the important ways to grasp the status of city road congestion. Video processing technology can process large amount of information, and easy to install without destruction of the ground. Because of these advantages, video processing technology has good application prospects in the field of traffic congestion identification of city road. Now, existed video processing technology which generally focused on road traffic incident detection or traffic parameter extraction has acquired some good results, but it hasn’t been used in the part of traffic congestion identification. Therefore, study on video-based traffic congestion identification technology of city road, from the aspect of efficiency and real-time, has important practical significance.This paper studied on the way of traffic congestion identification technology of city road based on video, including two aspects: One is the method of getting the traffic parameters by image processing, the other is the algorithm of traffic congestion identification technology.In the part of getting traffic parameters by image processing, the way of background modeling based on non-parametric kernel density algorithm was selected and the way of denoising by comparing probability was proposed. Then, Kalman filtering algorithm and the virtual test line were used to get the traffic parameters. The way of background modeling based on non-parametric kernel density algorithm has better results than other algorithm, and could model background for slow movement. The way of denoising by comparing probability can denoise caused by leaves jittering. The presented algorithm above can detect vehicle targets perfectly, and improve vehicle identification accuracy.In the part of traffic congestion identification technology, we improved the traditional fuzzy clustering algorithm. Firstly, the city road traffic congestion was made pre-identification by speed, Secondly, on the results of pre-identification, the traffic parameters weight sets were built at different traffic states. Finally, the result was getted by combining with pre-identification results, the actual identification results and historical identification results. The presented algorithm had better result and described the real traffic better, in which the traffic parameters in different traffic states have different importance and stability. Finally, the system of traffic congestion identification technology based on video was established using the video surveillance data in ChongQing. The experimental results show that this algorithm can get the traffic parameters more accurate, and the algorithm of traffic congestion identification has better sensitivity and lower percentage of two or more congestion level jumps. This system can realize the traffic congestion identification efficiently,and therefore improve the accuracy and feasibility.更多还原