【作者】 赵敏；

【导师】 孙棣华；

【作者基本信息】 重庆大学 ， 控制理论与控制工程， 2006， 硕士

【摘要】 发展智能公共交通系统是解决城市交通问题的有效途径。自动乘客计数技术是智能公交系统中的关键技术,是自动收集乘客上下车时间和地点的最有效方法之一,对实现公交实时调度,优化公交线路和进行交通预测等有重要意义。由于我国公交上下车客流密度较大,目前国外广泛使用的自动乘客检测技术应用于我国公交客流量统计精度不高,无法满足客流信息采集的需求。本文在介绍客流信息采集技术,深入分析现有自动乘客计数技术的应用现状基础上,提出将数字信号处理技术与数字图像处理技术应用于自动乘客计数技术中,由此建立基于多目标识别的自动乘客计数系统的系统框架,阐述系统功能模块以及系统实现的支撑技术。其中,数字信号处理技术与数字图像处理技术是实现基于多目标识别的自动乘客计数技术的前提与基础,多目标检测与跟踪计数算法是技术实现与应用的核心。针对乘客上下车图像序列的特点,在研究和比较传统的多目标检测与跟踪算法的基础上,提出和实现了适用于上下车乘客检测与跟踪计数的算法。在运动目标检测方面,给出了两步运动目标检测算法:首先提出了一种基于块平均灰度差值的自适应运动目标存在检测算法进行运动目标存在检测;其次利用有效的图像滤波、图像分割、形态学处理方法进行运动目标提取,并提出了一种简单实用的运动目标分割方法——基于头顶图像灰度值统计的半阈值分割结合自适应阈值分割的阈值分割算法,与其它经典的图像分割算法比较,具有计算简单、分割速度快的特点。在运动目标跟踪方面,设计的基于目标特征匹配的跟踪算法对质心欧氏距离代价函数进行了改进;并利用目标的运动特性,预测目标的运动位置,以缩小目标搜索匹配的范围;建立每个被跟踪目标的“目标链”,进而建立目标的关联关系,保证跟踪的稳定性和准确性。本文对现场采集的乘客上下车图像序列进行了大量实验,实验结果表明:基于多目标识别的自动乘客计数技术成功地实现了上下车乘客人数的自动统计,对上下车乘客人数的统计在90%以上的试验中,检测精度可达到90%以上。给出的多目标识别算法与其它算法相比,对多乘客目标进行检测与跟踪计数,在减少计算量、提高系统处理速度的同时,具有较高的精度。更多还原

【Abstract】 Developing intelligent public traffic system (IPTS) is an effective way to resolve traffic problems. In IPTS automatic passenger counting (APC) is a crucial technology, which can be used to automatically record the number of passengers boarding and alighting at individual bus stops. This paper introduces a new passenger counting technology, which is different from existing APC and is video based capable of addressing the need for automatically counting passengers on congested transit routes with high density passenger boarding and alighting activities. Accurate ridership information obtained from APC will help monitoring transit performance and improving the planning, scheduling and operations of the service.Based on the discussion of ridership information collection and the deeply analysis of existing APC technologies, the architecture of APC system based on multi-objects recognition is brought forward and analyzed in detail, including the module function and key technologies of system realization. Technologies of digital signal processing and digital image processing are the basis of APC, and multi-objects detecting and tracking algorithm is the essential factor to implement it.Comparing with traditional multi-objects detecting and tracking algorithm, a simple algorithm adapted to detect and track boardings and alightings is proposed. In term of moving objects detection, two-step moving objects detection algorithm is introduced. Firstly, a new algorithm of moving objects existing detection based on difference between blocks’average value of gray scale is presented to detect whether moving objects exist. Secondly, some effective approaches are adopted to extract moving objects such as image filtering, image segmentation, morphology processing etc. Besides, a kind of effective and simple thresholding methods for image segmentation is proposed, which adopts fixed threshold associated with adaptive threshold. In term of moving objects tracking, the feature-based tracking approach is adopted and a kind of cost function is improved in the tracking and counting algorithm. Besides, moving characteristics of targets are used to predict the search areas of matching objects. And then, moving state and feature value of objects in current frame are recorded to ensure the continuity of the dynamic tracking depended on object-chain.The APC technology based on multi-objects recognition was tested by capturing field image sequences of passengers getting on and off buses in Route 462, Chongqing,更多还原