{
  "paper_id": "3449722c0022407eca663e15c6981ddf",
  "paper_title": "A People Counting System Based on Dense and Close Stereovision",
  "paper_authors": [
    "Tarek Yahiaoui",
    "Cyril Meurie",
    "Louahdi Khoudour",
    "François Cabestaing"
  ],
  "publication_date": "",
  "publication_venue": "Image and Signal Processing",
  "abstract": "We present in this paper a system for passengers counting in buses based on stereovision. The objective of this work is to provide a precise counting system well adapted to buses environment. The processing chain corresponding to this counting system involves several blocks dedicated to the detection, segmentation, tracking and counting. From original stereoscopic images, the system operates primarily on the information contained in disparity maps previously calculated with a novel algorithm. We show that one can obtain a counting accuracy of 99% on a large data set including specific scenarios played in laboratory and on some video sequences shot in a bus during exploitation period.",
  "keywords": [
    "stereovision",
    "people counting",
    "passenger counting",
    "disparity map",
    "dense stereo matching",
    "bus environment"
  ],
  "paper_institutions": [
    "French National Institute for Transport and Safety Research (INRETS-LEOST)",
    "University of Sciences and Technology of Lille (LAGIS laboratory, UMR 8146)"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:47:39.454269Z",
  "template_tag": "simple",
  "summary": "### 摘要\n本文提出了一种基于密集立体视觉的公交车乘客计数系统。针对公共交通运营中对精确计数信息的需求，以及现有红外、超声等计数系统在拥挤场景下误差率高的问题，系统采用近距离立体视觉技术，通过改进的SAD立体匹配算法计算视差图，再转化为高度图进行分割、跟踪和计数。处理链包括检测、分割、跟踪和计数四个模块，传感器垂直安装在公交车门上方以减少遮挡。系统在包含30个实验室特定场景（分日光和人工光两类）和3段真实公交车视频的数据集上进行了评估，整体计数精度达到99%，误差率约为1%。结果表明，该方法能应对高密度人群、反向移动、不同身高及携带行李等复杂情况，但存在轻微低估倾向，主要原因是小尺寸人员难以检测、分割结构元素大小和轨迹合并问题。未来将使用包含150个真实公交车序列的新数据集进行更密集的评估，期望误差保持在2%以下。核心机制是通过引入四种相似性约束（中心灰度、边缘类型、CCN灰度轮廓、区域运动）对SAD差异度量进行加权，其公式为C(x,y,s)=(α+β+γ+μ)Σ|G(x+i+s,y+j)-D(x+i,y+j)|，从而提高了遮挡区域的匹配精度。\n\n### 关键要点\n- 系统基于密集立体视觉，传感器垂直固定在公交车门上方，以降低遮挡并适应乘客头部检测。\n- 处理链包含四个模块：检测（立体匹配）、分割、跟踪和计数，构成完整流程。\n- 提出改进的SAD算法，在原有差异度量上引入四种相似性约束，并通过加权系数α、β、γ、μ进行调节。\n- 四种约束分别对应中心灰度相似性、像素类型（是否边缘）、CCN灰度轮廓相似性以及区域是否运动，当约束满足时系数取小值（0<系数<1），否则为1。\n- 与SAD、SSD、ZSAD、ZSSD等经典算法相比，该方法的匹配错误率降低了3%，尤其在遮挡区域效果显著。\n- 视差图通过三角测量转换为高度图，基于高度区间进行分割，并应用圆形结构元素的开运算提取头部内核。\n- 跟踪阶段使用卡尔曼滤波预测头部位置，并结合属性向量（大小、宽度、平均高度、灰度、坐标等）进行匹配。\n- 计数方法基于有效轨迹分析，判断乘客的进出行为。\n- 评估数据集包含30个实验室场景（1-15日光、16-30人工光）和3段真实公交车视频，场景多样且难度各异。\n- 实验室场景中，计数误差范围在-1到+1之间，真实公交车场景前两段无误差，第三段有轻微低估。\n- 整体错误率约为1%，计数精度达99%，验证了系统在复杂情况下的鲁棒性。\n- 系统倾向低估人数，原因包括小尺寸人员检测困难、分割结构元素大小设置不当以及两条轨迹合并。\n- 未来计划使用150个真实立体视频序列进行更深入评估，以进一步验证系统的可靠性。",
  "available_templates": [
    "deep_read",
    "simple"
  ]
}