{
  "paper_id": "b98a473fef31cb76b6ac4b26027dc88e",
  "paper_title": "Counting people by RGB or depth overhead cameras",
  "paper_authors": [
    "Luca Del Pizzo",
    "Pasquale Foggia",
    "Antonio Greco",
    "Gennaro Percannella",
    "Mario Vento"
  ],
  "publication_date": "",
  "publication_venue": "Pattern Recognition Letters",
  "abstract": "In this paper we present a vision based method for counting the number of persons which cross a virtual line. The method analyzes the video stream acquired by a camera mounted in a zenithal position with respect to the counting line, allowing to determine the number of persons that cross the virtual line and providing the crossing direction for each person. The proposed approach has been specifically designed to achieve high accuracy and computational efficiency, so as to allow its adoption in real scenarios. An extensive evaluation of the method has been carried out taking into account the main factors that may impact on the counting performance and, in particular, the acquisition technology (traditional RGB camera and depth sensor), the installation scenario (indoor and outdoor), the density of the people flow (isolated people and groups of persons), the acquisition frame rate, and the image resolution. We have also analyzed the combination of the outputs obtained from the RGB and depth sensors as a way to improve the counting performance. The experimental results confirm the effectiveness of the proposed method, especially when combining RGB and depth information, and the tests over three different CPU architectures demonstrate the possibility of deploying the method both on high-end servers for processing in parallel a large number of video streams and on low power CPUs as those embedded on commercial smart cameras.",
  "keywords": [
    "People counting",
    "Overhead camera",
    "Video analytics",
    "Retail analytics"
  ],
  "paper_institutions": [
    "University of Salerno"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:46:25.832638Z",
  "template_tag": "simple",
  "summary": "### 摘要\n本文提出一种基于视觉的顶置相机人流量统计方法，用于统计穿越虚拟线的人数并提供穿越方向。方法由前景检测和计数传感器两阶段组成，通过背景差分和动态背景更新检测运动目标，并利用基于条纹激活序列的传感器进行计数，避免使用目标检测与跟踪，从而提升鲁棒性和计算效率。在受控数据集（Dataset_C）和不受控数据集（Dataset_U）上进行了广泛实验，涵盖RGB和深度传感器、室内外场景、不同人流密度、帧率（5-30fps）和分辨率（160×120至640×480）。采用精确率、召回率和F值作为指标，结果表明深度传感器优于RGB，尤其在群组场景；将RGB与深度输出通过最大规则融合可进一步提升性能并平衡误报与漏报。在处理器性能测试中，方法在PC、ARM和MIPS平台上均能实时运行，低分辨率下可达到高帧率，验证了其在服务器端和嵌入式设备上的部署可行性。核心机制是使用动态背景更新公式 B_t = α*I_t + (1-α)*B_{t-1} 和基于条纹激活序列的计数逻辑，通过阈值θ_C和θ_K控制灵敏度。\n\n### 关键要点\n- 方法仅需前景检测，无需人员检测或跟踪，通过虚拟传感器直接计数，降低了复杂度和对前景错误的影响。\n- 背景更新采用指数平滑，更新率降低到10%，以大幅减少计算负担，同时保持性能稳定。\n- 传感器由宽度W和高度H的矩形区域构成，分为K条条纹，每条分为两个单元，通过单元激活序列判断穿越方向和人数。\n- 参数设置中，高度H设为平均人体深度的两倍，阈值θ_C=0.2，θ_K=3，在独立验证序列上调优。\n- 在Dataset_C上，RGB传感器的孤立穿越F值为0.990，群组降至0.908；深度传感器在两种密度下F值分别达0.990和0.976，性能稳定。\n- 将RGB与深度输出通过最大规则组合，在Dataset_C上整体F值提升至0.991，优于单一传感器。\n- 在不受控数据集Dataset_U上，深度传感器仍显著优于RGB，尤其群组场景F值从0.810提升至0.926。\n- 分辨率从640×480降至320×240时，RGB和深度传感器的F值分别下降约2%和1%；降至160×120时损失约4-5%。\n- 帧率低于15fps时性能急剧下降，30-15fps之间保持稳定，低帧率下无法正确激活传感器序列。\n- 与基于跟踪的方法相比，本方法在RGB和深度数据上均明显提升精确率和召回率，证明避免跟踪的鲁棒性。\n- 计算分析显示，在PC上以640×480分辨率处理可达189fps，ARM平台17fps，MIPS平台仅5fps；但降到320×240后MIPS可达23fps，满足实时需求。\n- 预处理的引入（3×3高斯滤波）减少了噪声导致的虚警和漏检，相比未加预处理的版本在RGB和深度上F值均有提升。"
}