道路人流量状态监测

发布时间：2018-01-18 09:50

本文关键词：道路人流量状态监测　出处：《贵州民族大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着社会发展和城镇化加速推进,群体性社会活动逐渐增多,例如运动会、音乐会、宗教集会等。在类似的场景中,由于人流之间运动快慢和方向不一,局部区域的人群密度不同,加上建筑通道承载量的客观限制,使得在群体性活动中极易发生人群拥堵,甚至踩踏事故,造成生命财产损失。为有效地避免群体性活动中意外事故的发生,智能监控技术已经被广泛地应用于车站、机场、地铁以及大型的购物中心等领域。而在群体性智能监控中,人群密度估计是后续状态监测的核心技术,也是计算机视觉和图像处理领域的研究热点。为了对群体进行有效的人流量状态监测,提高道路的利用率,避免公民不必要的生命财产损失,提出了一种道路人流量状态监测方法,主要包括运动信息提取、人群前景目标检测、运动流分割和人流量状态识别。开展的具体工作如下:(1)采用全局光流法获取场景的运动信息,包括运动方向和快慢;(2)借鉴粒子动力学的理论方法实现前景目标检测;(3)以运动前景为掩模提取对应帧的光流角度,再利用动态C均值聚类方法实现前景运动流分割;(4)在光流场中计算不同运动流的前景光流面积特征和基于光流角度的能量、熵、均匀度、对比度等纹理特征,并以此作为样本的特征向量,建立多元线性回归模型,实现人群密度估计,进而达到对人流量状态监测的目的。(5)将建议算法与自组织神经网络方法进行测试比较,对比验证本文算法的优越性。通过对比实验证实,本文算法切实可行,为后续行为分析和理解提供了重要的参考价值。
[Abstract]:With the social development and the accelerating urbanization, group social activities gradually increased, such as games, concerts, and other religious gatherings. In a similar scenario, due to the flow of people between the speed of movement and direction of a different population density in local area, building channels plus bearing objective restrictions, so prone to the crowd congestion in group activities, and even a stampede, resulting in the loss of life and property. In order to effectively avoid accidents occurred in group activities, intelligent monitoring technology has been widely used in the station, airport, subway, shopping malls and other fields. In the group of intelligent monitoring, crowd density estimation is the core technology of state monitoring of follow-up, it is also a research hotspot in computer vision and image processing. In order to carry out effective monitoring the flow of people in groups, improve the utilization of the road to avoid Citizens of the unnecessary loss of life and property, proposed a road traffic monitoring method, including motion information extraction, the crowd foreground object detection, motion segmentation and traffic state recognition. The specific work is as follows: (1) the global optical flow method to obtain the motion information of the scene, including the movement direction and speed; (2) from the theory of particle dynamics for foreground object detection; (3) to the foreground mask for the extraction of the corresponding frame optical flow angle, using dynamic C clustering method to realize the moving foreground segmentation; (4) the prospect of optical flow calculation of different flow in the area features in the optical flow field and optical flow of energy, based on entropy, uniformity, contrast and texture features, and as a feature vector of sample, establish multiple linear regression model, the estimated density of the crowd, so as to the traffic condition monitoring purposes. (5) comparing the proposed algorithm with the self-organizing neural network, the superiority of the algorithm is verified by comparison. Compared with experiments, the algorithm is feasible and provides an important reference value for subsequent behavior analysis and understanding.

【学位授予单位】：贵州民族大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】