{
  "paper_id": "0868433539393c955e60b35b829a7964",
  "paper_title": "Person Counting Using Stereo",
  "paper_authors": [
    "David Beymer"
  ],
  "publication_date": "",
  "publication_venue": "Proceedings Workshop on Human Motion",
  "abstract": "Stores and shopping malls would like to keep track of shopper volume by employing automatic techniques for counting shoppers. Existing approaches instrument doors with infrared beams and count beam interruptions, but this approach cannot resolve groups of people well. We are applying a vision-based approach that detects and tracks people from a stereo camera mounted above a door and pointing down. After applying real-time stereo and 3D reconstruction, the system segments the scene by selecting stereo pixels falling inside a 3D volume of interest, which is placed to capture the heads and torsos of adult shoppers. The main novelties of our approach include (1) remapping the stereo disparities to an orthographic \"occupancy map\", which simplifies person modeling, and (2) tracking people using a Gaussian mixture model. On a test set of 900 enter/exit events in 4 hours of video, our system has achieved a net counting error rate of just 1.4%.",
  "keywords": [
    "person counting",
    "stereo vision",
    "occupancy map",
    "Gaussian mixture model",
    "tracking",
    "volume of interest"
  ],
  "paper_institutions": [
    "SRI International"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:46:11.212538Z",
  "template_tag": "simple",
  "summary": "### 摘要\n该论文针对商店和购物中心对顾客流量计数的需求，提出了一种基于立体视觉的人员计数系统。现有红外光束计数方法无法有效分辨并排行走的人群，而论文采用安装在门上方、俯视拍摄的立体摄像机，通过实时立体匹配和三维重建，利用三维感兴趣区域（VOI）分割出头和躯干区域。主要创新包括将视差重投影为正交视角的“占用图”（occupancy map），简化了人员建模，以及使用高斯混合模型进行多人跟踪。实验在SRI校园双门入口处采集约4小时视频数据，共905个进出事件，系统净计数误差仅为1.4%。核心机制是用竞争性高斯混合模型拟合占用图，通过迭代分割和参数更新实现人员跟踪。系统运行帧率为20-25 Hz，在400 MHz Pentium II上实现。未来工作包括更多现场测试、减少高个子在占用图中多峰造成的过计数，以及简化标定流程。该系统适用于商店、商场等实际场景，但受限于安装高度（8-9英尺）需广角镜头，并需处理严重径向畸变。\n\n### 关键要点\n- 人员计数问题对于商场市场信息和租赁率评估很重要，现有红外光束计数器无法解决并排人群的欠计数问题。\n- 采用立体摄像头垂直向下安装，利用3D感兴趣区域（VOI）滤除无关物体，如购物车和儿童。\n- 视差通过查表法实时映射到正交占用图，每个像素贡献权重为disp_nom/disp以补偿深度。\n- 占用图将人的头肩投影为近似高斯形状，便于用高斯混合模型表示。\n- 比较了加性模型和竞争性高斯混合模型，发现竞争性模型能更好地区分并排人群。\n- 跟踪采用卡尔曼滤波预测位置，然后通过迭代分割和牛顿-拉夫森更新参数拟合占用图。\n- 新人员的检测基于残差图(occ-G)的局部最大值，并设置阈值避免重复检测。\n- 轨道退出条件包括位置出界、高斯宽度过小或强度过弱，随后分类为进入/退出/重进/经过。\n- 实验使用约3.5小时食堂和30分钟礼堂视频，人工标注地面真值，计算净计数误差、进入误差和退出误差。\n- 总净计数误差为1.4%（13/905），其中Cafe1为2.1%，Cafe2为0.9%，All hands为0%。\n- 主要误差来源是高个子在占用图中呈现多峰导致过计数，以及轨道退出时的异常运动导致分类拒绝。\n- 系统在400 MHz双Pentium II上单线程实现，帧率20-25 Hz，实时性能满足应用需求。"
}