{
  "paper_id": "75a799deaf75198cfd2bacd220521db6",
  "paper_title": "Real-Time Human Pose Recognition in Parts from Single Depth Images",
  "paper_authors": [
    "Jamie Shotton",
    "Toby Sharp",
    "Alex Kipman",
    "Andrew Fitzgibbon",
    "Mark Finocchio",
    "Andrew Blake",
    "Mat Cook",
    "Richard Moore"
  ],
  "publication_date": "2011",
  "publication_venue": "Proceedings of the 2011 Conference on Computer Vision and Pattern Recognition (CVPR)",
  "abstract": "We propose a new method to quickly and accurately predict human pose—the 3D positions of body joints—from a single depth image, without depending on information from preceding frames. Our approach is strongly rooted in current object recognition strategies. By designing an intermediate representation in terms of body parts, the difficult pose estimation problem is transformed into a simpler per-pixel classification problem, for which efficient machine learning techniques exist. By using computer graphics to synthesize a very large dataset of training image pairs, one can train a classifier that estimates body part labels from test images invariant to pose, body shape, clothing, and other irrelevant ones. Finally, we generate confidence-scored 3D proposals of several body joints by reprojecting the classification result and finding local modes. The system runs in under 5ms on the Xbox 360. Our evaluation shows high accuracy on both synthetic and real test sets, and investigates the effect of several training parameters. We achieve state-of-the-art accuracy in our comparison with related work and demonstrate improved generalization over exact whole-skeleton nearest neighbor matching.",
  "keywords": [
    "human pose estimation",
    "depth images",
    "body part classification",
    "randomized decision forests",
    "mean shift clustering",
    "Kinect",
    "real-time"
  ],
  "paper_institutions": [
    "Microsoft Research, Cambridge, UK",
    "Xbox Incubation",
    "ST-Ericsson"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:47:21.813220Z",
  "template_tag": "simple",
  "summary": "### 摘要\n本论文提出了一种从单个深度图像实时且准确预测人体姿态（即身体关节的三维位置）的新方法，无需依赖前序帧信息。方法借鉴了物体识别策略，通过设计一个身体部位中间表示，将困难的姿态估计问题转化为逐像素分类问题，并采用高效的机器学习技术。利用计算机图形学合成大规模训练图像对，训练出的分类器能够对测试图像的身体部位标签进行估计，且对姿态、体型、衣着等因素具有不变性。最终，通过将分类结果重投影并寻找局部模式，生成多个身体关节的置信度加权的三维候选位置。系统在Xbox 360上运行时间低于5毫秒。实验在合成和真实测试集上均展现了高精度，并研究了多个训练参数的影响。与相关工作相比，我们取得了最先进的精度，并证明比精确的全身最近邻匹配具有更好的泛化能力。\n\n### 关键要点\n- 方法核心是将姿态估计转化为逐像素的身体部位分类问题，利用中间身体部位表示，避免了对关节组合的搜索。\n- 使用深度比较特征，通过归一化偏移保证深度不变性，特征计算简单，适合GPU实现。\n- 采用随机决策森林进行分类，训练数据由合成深度图像生成，包含大量姿态、体型、衣着和相机变化的样本。\n- 合成训练数据基于运动捕捉数据库，经过最远邻聚类去冗余，选取100,000个姿态，保证多样性。\n- 数据生成管线包含15个基础角色模型、随机梅花发型服装、身高体重变化、相机随机位置和噪声模拟。\n- 关节候选位置通过均值漂移密度模态估计获得，结合像素权重和表面面积归一化，并学习每部分带宽和偏移参数。\n- 实验设置：3棵树、深度20、每棵树300k训练图像、每图像2000个样本像素，训练在1000核集群上约一天。\n- 分类精度随训练图像数量对数增长，在约100k图像后趋于饱和；树深度影响最大，900k训练集可避免过拟合。\n- 在合成测试集上，均值漂移聚类达到0.731 mAP，快速聚类为0.677 mAP；在真实数据集上与Ganapathi等人相比精度更高且速度快至少10倍。\n- 与最近邻匹配相比，本方法在达到约150k训练图像前，优于甚至理想的骨架匹配，且能自然处理裁剪和遮挡。\n- 在360°旋转场景下mAP为0.655，表明分类器能学习区分前后朝向；多人的情况也能处理。\n- 方法的局限包括对交叉手臂等细微深度变化可能失效，但通过置信度门控可保持高精度。"
}