【作者】 黄波；

【导师】 赵敏； 唐毅；

【作者基本信息】 重庆大学 ， 工程硕士（控制工程领域）（专业学位）， 2017， 硕士

【摘要】 夜间高速公路相比白天更容易发生交通事件,导致交通拥堵。随着视频监控系统在高速公路的大范围普及应用,基于视频的高速公路拥堵事件检测得到迅速发展。目前基于视频的拥堵状态判别算法大都适用于光线充足的白天场景,由于夜间高速公路存在照明条件差、环境相对较暗、车身基本不可见等特点,使得现有基于视频的拥堵判别算法难以满足夜间高速公路场景拥堵检测需求。因此,充分利用高速公路现有条件,设计与实现针对夜间高速公路场景的拥堵状态判别算法具有重要的理论和实际意义。本文针对夜间高速公路这一特定场景,在分析了基于视频的夜间高速公路拥堵状态判别问题及难点后,重点研究了夜间行进车辆车灯的提取、适用于夜间的交通参数获取以及夜间拥堵判别模型的建立,最终形成一套基于视频的夜间高速公路拥堵状态判别算法。主要研究内容包括:(1)在车灯提取方面,首先针对夜间视频图像存在的大量噪声问题,运用图像增强的方法进行预处理操作,在降低噪声干扰的同时,增强车灯与背景区域的对比度。然后针对传统的车灯分割方法在夜间高速公路场景下适应性较差的问题,采用基于遗传算法的最大熵双阈值的车灯分割方法。最后针对分割后的图像存在地面反射光干扰的问题,提出基于梯度信息的反射光消除方法。实验结果表明,该方法有效地提高了车灯提取的准确性,同时为准确获取交通参数奠定了基础。(2)在夜间交通参数获取方面,针对在夜间高速公路场景下,由于无光照、车身不可见等导致常用的交通参数难以获取问题,利用高速公路的结构化特点,通过对道路的车道分界线提取来获取该路段的长度及车道数等道路信息,在此基础上结合车灯与道路信息实现夜间高速公路场景下平均交通密度和时间占有率的获取。(3)在夜间拥堵判别方面,针对各交通状态间的模糊性关系,通过分析平均交通密度和时间占有率两个交通参数与道路拥堵间的关系,利用FCM算法对夜间交通拥堵状态进行判别,并通过对投票机制的方法降低误检率,提高了夜间拥堵判别的准确性。最后,综合上述研究成果,形成了一套基于视频的夜间高速公路拥堵事件检测方法,并在VS2010和OpenCV2.4.8平台下进行设计与实现,利用实地采集的高速公路的夜间视频数据进行实验验证。实验结果表明,论文方法能够较准确的提取车灯目标,且在保证算法实时性的前提下,提高了实际场景下夜间拥堵状态判别的准确性。更多还原

【Abstract】 The highway at night compared to daytime more prone to traffic incident,causing traffic congestion.With the popularization and application of video monitoring system in a wide range of highway,highway congestion event detection based on video has been developing rapidly.The current video discrimination algorithm based on the congestion state is only applicable to broad daylight scenes,at night,there are many disadvantages such as poor lighting conditions,relatively dark environment and invisible body,the existing congestion identification algorithm of based on video is not difficult to meet the demand of highway congestion detection of night scene.Therefore,it is of great theoretical and practical significance to make full use of the existing conditions of highway and to design and implement the congestion state estimation algorithm.In view of this particular scene at night highway,in the analysis of the problems and difficulties of congestion distinguishing based on video at night highway state,this paper focuses on the research of the extraction of vehicle lights at night,the acquisition of traffic parameters suitable for the night,and the establishment of the model of nighttime congestion,finally form a traffic congestion discriminant algorithm which based on video on highway at night.The main research contents include:(1)In terms of lamp extraction,firstly,for the problems of there are a lot of noise in video images at night,by using the method of image enhancement preprocessing operations,to reduce noise interference and enhance the contrast of light and background area.Then,for the problems of poor adaptability of the conventional vehicle lights segmentation at night scene,using maximum entropy method of double threshold based on genetic algorithm for lamp segmentation.Finally,for the problems of the interference of ground reflected light which after image segmentation,a method of reflection light elimination based on gradient information is proposed.The experimental results show that,this method can effectively improve the accuracy of vehicle light extraction,while laying the foundation for accurate access to traffic parameters.(2)In terms of nighttime traffic parameters acquisition,in the night scene of highway,because there is no light,the body of car is not visible as a result of general traffic parameters are difficult to obtain,utilize the structural features of highway,to obtain the road length and number information through extraction the road lane line,based on this,get to combine with light and road information to achieve an average traffic density and time occupancy ratio of the highway scenes at night.(3)In terms of nighttime congestion identification,based on the fuzzy relationship between traffic states,by analysis of the relationship between two traffic parameters which average traffic density and time occupancy ratio,use the FCM algorithm to judge the traffic congestion at night,and reduce the error rate by the method of voting mechanisms,improved the accuracy of congestion identification at nighttime.Finally,according to the above research,formed a set of congestion event detection method based on video at nighttime highway,and to design and implementation under the platform of VS2010 and OpenCV2.4.8,and experiments were carried out using night video data collected on the spot of highway.The experimental results show that,the method can accurately extract the target of the lamp,and in the premise of real-time algorithm,improved the accuracy of congestion state discriminant at night under the actual scene.更多还原