{
  "paper_id": "a93a3f6a818ba29ddb4b8792ed4114de",
  "paper_title": "People counting via multiple views using a fast information fusion approach",
  "paper_authors": [
    "Mikaël A. Mousse",
    "Cina Motamed",
    "Eugène C. Ezin"
  ],
  "publication_date": "2016-02-09",
  "publication_venue": "Springer Science+Business Media New York",
  "abstract": "Real-time estimates of a crowd size is a central task in civilian surveillance. In this paper we present a novel system counting people in a crowd scene with overlapping cameras. This system fuses all single view foreground information to localize each person present on the scene. The purpose of our fusion strategy is to use the foreground pixels of each single view to improve real-time objects association between each camera of the network. The foreground pixels are obtained by using an algorithm based on codebook. In this work, we aggregate the resulting silhouettes over cameras network, and compute a planar homography projection of each camera's visual hull into ground plane. The visual hull is obtained by finding the convex hull of the foreground pixels. After the projection into the ground plane, we fuse the obtained polygons by using the geometric properties of the scene and on the quality of each camera detection. We also suggest a region-based approach tracking strategy which keeps track of people movements and of their identities along time, also enabling tolerance to occasional misdetections. This tracking strategy is implemented on the result of the views fusion and allows to estimate the crowd size dependently on each frame. Assessment of experiments using public datasets proposed for the evaluation of counting people system demonstrates the performance of our fusion approach. These results prove that the fusion strategy can run in real-time and is efficient for making data association. We also prove that the combination of our fusion approach and the proposed tracking improves the people counting.",
  "keywords": [
    "Visual surveillance",
    "People counting",
    "Homography",
    "Data association",
    "Tracking",
    "Overlapping cameras"
  ],
  "paper_institutions": [
    "Université d'Abomey-Calavi",
    "Université du Littoral Côte d'Opale"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:47:07.350897Z",
  "template_tag": "simple",
  "summary": "### 摘要\n本文针对视频监控中人群计数的实时性问题，提出了一种基于多视角快速信息融合的多人计数系统。该系统利用重叠摄像头网络，通过码本模型检测各视图的前景像素，并用凸包算法将前景区域近似为多边形。这些多边形经平面单应性投影到地面后，融合策略依据场景几何属性与每个摄像头检测质量进行最优视图选择，同时处理动态遮挡和误检。此外，提出了一种基于区域的跟踪策略，通过一致性和身份指示器维持人员身份，容忍短暂丢失，并结合融合结果实现逐帧人数估计。在EPFL和PETS 2009/2010公共数据集上的实验表明，该方法在Recall、Precision和F-measure指标上优于或接近现有方法，处理速度比传统体素方法快，且无需监督学习，可在普通笔记本电脑上实时运行。未来将利用运动方向轨迹进行人群细分和事件识别。\n\n### 关键要点\n- 提出一种多视角人群计数系统，通过融合各视图前景信息定位场景中的每个人。\n- 前景检测采用基于码本模型的运动检测算法，具有高准确率和鲁棒性。\n- 使用凸包算法将前景像素近似为多边形，减少数据处理量，并考虑空洞和间断区域。\n- 通过平面单应性投影将每个摄像头的视觉外壳多边形映射到地面平面，实现跨视角关联。\n- 融合策略基于场景几何属性和检测质量，分三种情况处理多边形关联：无关联、一对一关联、一对多关联，并采用射线投射算法判断点是否在多边形内。\n- 融合过程优先选择检测到最低点（最接近地面）的摄像头视图，若无法区分则比较最高顶点与最低顶点投影距离。\n- 动态遮挡情况（一个多边形关联多个多边形）下，选择关联多边形最多的摄像头视图。\n- 跟踪策略扩展了单摄像机算法，采用一阶位置预测、空间验证门、颜色直方图（每通道32 bin）和Bhattacharya距离进行数据关联。\n- 一致性和身份指示器递归更新，允许容忍短暂丢失，超过3秒零值则终止轨迹。\n- 计数策略在跟踪器检测到新对象时计数加一，同时将暂时丢失的对象计入当前帧人数。\n- 在EPFL Terrace1和PETS 2009序列上测试，PETS序列人数高达40人，场景存在光照变化、阴影和频繁遮挡。\n- 实验结果显示，在Time_12-34_S0上Recall=0.98, Precision=0.99, F-measure=0.98；在Time_12-34_S2上三项均为0.98；在Terrace1上Recall=0.98, Precision=0.99, F-measure=0.98，优于或接近比较方法。\n- 处理速率在2.02至7.97帧/秒之间，随场景人数和摄像头数量变化，验证了实时性。"
}