{
  "paper_id": "a95c5a20bd2c6a9efcb86276305d049a",
  "paper_title": "People-Flow Counting Using Depth Images for Embedded Processing",
  "paper_authors": [
    "Guilherme S. Soares",
    "Rubens C. Machado",
    "Roberto A. Lotufo"
  ],
  "publication_date": "",
  "publication_venue": "Image Analysis and Recognition",
  "abstract": "This paper proposes a people-flow counting algorithm using top-view depth images for implementation on low-power, embedded processors. In the people detection stage the algorithm uses morphological connected filters to find head candidates, and in the tracking stage it uses Kalman filtering in order to obtain good predictions in frames where detection fails. A fast interpolation algorithm is also proposed, which estimates the values of pixels affected by noise and generates an image with a continuous domain. The experiments were done using a Kinect sensor and the processing was performed in real time on a Raspberry Pi 3. The dataset consisted of 4025 short video sequences of people entering and exiting indoor environments, obtained from three different installations. The algorithm proved to be adequate for an embedded application, reaching an accuracy of 98% for frame rates as low as 5.5 FPS.",
  "keywords": [
    "People counting",
    "Depth images",
    "Embedded processing"
  ],
  "paper_institutions": [
    "School of Electrical and Computer Engineering, University of Campinas, Campinas, SP, Brazil"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:47:24.698998Z",
  "template_tag": "simple",
  "summary": "### 摘要\n本文针对低功耗嵌入式处理器上的人员流量计数问题，提出了一种基于俯视深度图像的算法。算法包括预处理、人员检测、跟踪和计数四个阶段。预处理阶段采用下采样、深度映射、反转和插值算法修复噪声像素，其中提出了一种利用距离变换和排序的快速插值方法。人员检测阶段结合形态学h-maxima算子、阈值化、面积开运算和区域属性（最小面积和长宽比）筛选头部候选。跟踪阶段使用卡尔曼滤波器预测和校正每个目标的位置，并处理漏检和帧丢失问题，通过动态调整距离阈值适应处理帧率变化。实验使用Kinect传感器在三个不同室内安装场景采集了4025个短视频序列，在Raspberry Pi 3上实时运行，平均帧率约5.5 FPS，最佳准确率达到98%。实验表明系统准确性受图像有效像素比例影响，整体算法适用于资源受限的嵌入式应用。未来可扩展跟踪匹配特征（如形状、面积）以提升鲁棒性。\n### 关键要点\n- 算法专为嵌入式低功耗环境设计，在Raspberry Pi 3上实现实时处理，平均帧率仅5.5 FPS即可获得高精度。\n- 采用俯视深度图像，利用Kinect传感器捕获10-bit深度数据，映射为8-bit线性图像并反转。\n- 提出快速插值算法：将无效像素置零，通过距离变换计算每个无效像素到最近有效像素的距离，按距离排序后依次分配有效邻居的最大值。\n- 人员检测使用形态学h-maxima获取有意义的区域最大值，并通过阈值化和面积开运算去除噪声和无效对象。\n- 检测阶段筛选区域最大值需满足最小面积和最小长宽比，从而定位头部候选，计算质心作为跟踪输入。\n- 跟踪阶段为每个质心建立卡尔曼滤波器，预测下一帧位置，并与当前检测质心计算欧氏距离，按距离升序匹配。\n- 匹配条件：距离小于阈值d_max，且质心未与其他已匹配质心冲突；未匹配的检测质心初始化为新轨迹，未匹配的预测质心可传播直到达到最大次数。\n- 采用六维状态向量（位置和速度）的恒速模型，加速度由过程噪声建模；观测模型仅测量位置。\n- 帧丢失时，预测阶段重复n次（n = ΔT/Δt），且最大匹配距离d_max = v_max·ΔT随帧间隔增大而增加。\n- 实验使用三个不同安装场景，传感器高度从2.90m到4.00m，门类型、光照条件等各异，数据集共4025个短视频。\n- 结果按有效像素率排序，安装C的准确率最高（98%），安装B因有效像素率低（66.8%）准确率降至84%，表明图像质量对性能影响显著。"
}