基于视频的公交客流统计方法研究

发布时间：2018-07-11 15:21

本文选题：乘客人数统计 + 目标检测　；参考：《重庆大学》2014年硕士论文

【摘要】：公交客流信息能辅助作出合理的发车计划，更科学的调度规划，使现有公交资源得以更加充分的利用，获得最佳的运营效果。目前，常用的客流统计技术并不适用于我国城市公交客流量大，乘客上下车过程拥挤无序的现状，且无法得到准确的统计结果。本文针对基于单目视觉的公交乘客人数统计结果不准确、判定准则不稳定的现状，结合公交车门附近乘客运动行为的复杂性和多样性，提出一种基于视频的乘客多运动行为分析客流计数方法。结合公交环境下乘客图像序列的特点，文中采用了一种两步骤的运动目标检测方法。即首先用四帧间差分相乘算法对图像序列进行初检测，初步定位乘客目标区域，，然后利用基于人体头部外形特征的霍夫圆(Hough Circle)检测和结合人头灰度特征对帧差结果进行检测，筛选出乘客头部。针对公交站点车门附近乘客运动行为的复杂性和多样性，及乘客运动行为对计数判定准则的干扰，在分析乘客运动行为的基础上提出一种新的计数判定准则。文中采用轨迹聚类的方法对乘客运动行为进行分析。结合轨迹的空间特征和方向特征改进相似性测量方法进行轨迹距离的计算，并采用层次聚类方法进行聚类。分析聚类结果中每一类别所对应乘客类的运动行为，并讨论各乘客类的运动行为对现有计数判定准则的影响，由此提出一种基于乘客运动行为分析的公交车客流计数判定准则。通过对乘客上下公交车图像序列进行的实验表明，基于视频的乘客多运动行为分析公交车客流统计方法能有效地完成乘客上下车计数统计，且本文的计数判定准则同其他的判定准则相比，本文的方法具有较高和较稳定的计数精度。
[Abstract]:Bus passenger flow information can help to make a reasonable departure plan, more scientific scheduling planning, so that the existing public transport resources can be more fully utilized, and get the best operation effect. At present, the commonly used statistical techniques of passenger flow are not suitable for the current situation of large traffic volume of public transport in our country, and the process of passengers getting up and down is crowded and disordered, and the accurate statistical results can not be obtained. Aiming at the inaccurate statistical results of bus passenger numbers based on monocular vision and the unstable judging criteria, this paper combines the complexity and diversity of passenger movement behavior near bus gates. A method of passenger flow counting based on video-based multi-motion behavior analysis is proposed. Combined with the characteristics of passenger image sequence in public transport environment, a two-step moving target detection method is adopted in this paper. First, the image sequence is initially detected by using the differential multiplication algorithm between four frames, and the passenger target area is preliminarily located. Then, the Hough Circle based on the human head shape feature is used to detect the frame difference result, and the human head gray level feature is combined to detect the frame difference result. Screen out the heads of the passengers. In view of the complexity and diversity of passenger's motion behavior near bus station door and the interference of passenger's motion behavior to counting criterion, a new counting criterion is proposed based on the analysis of passenger's motion behavior. In this paper, the trajectory clustering method is used to analyze the movement behavior of passengers. Combining the spatial and directional features of the trajectory, the similarity measurement method is improved to calculate the trajectory distance, and the hierarchical clustering method is used to cluster the trajectory distance. This paper analyzes the movement behavior of each passenger class in the clustering results, and discusses the influence of the movement behavior of each passenger class on the existing counting criteria, and puts forward a bus passenger flow counting criterion based on passenger movement behavior analysis. The experiments on the image sequences of passengers from upper and lower buses show that the statistical method of passenger flow analysis based on video multi-movement behavior can effectively complete the counting statistics of passengers on and off the bus. Compared with other criteria, the method presented in this paper has higher and more stable counting accuracy.
【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】：U491.17;U492.413

【参考文献】