{
  "paper_id": "e958b3f148db391a093940da98a573cd",
  "paper_title": "Temporal Convolutional Networks for Multiperson Activity Recognition Using a 2-D LIDAR",
  "paper_authors": [
    "Fei Luo",
    "Stefan Poslad",
    "Eliane Bodanese"
  ],
  "publication_date": "",
  "publication_venue": "IEEE Internet of Things Journal",
  "abstract": "Motion trajectories contain rich information about human activities. We propose to use a 2-D LIDAR to perform multiple people activity recognition simultaneously by classifying their trajectories. We clustered raw LIDAR data and classified the clusters into human and nonhuman classes in order to recognize humans in a scenario. For the clusters of humans, we implemented the Kalman filter to track their trajectories which are further segmented and labeled with corresponding activities. We introduced spatial transformation and Gaussian noise for trajectory augmentation in order to overcome the problem of unbalanced classes and boost the performance of human activity recognition (HAR). Finally, we built two neural networks, including a long short-term memory (LSTM) network and a temporal convolutional network (TCN) to classify trajectory samples into 15 activity classes collected from a kitchen. The proposed TCN achieved the best result of 99.49% in overall accuracy. In comparison, the TCN is slightly superior to the LSTM network. Both the TCN and the LSTM network outperform the hidden Markov model (HMM), dynamic time warping (DTW), and support vector machine (SVM) with a wide margin. Our approach achieves a higher activity recognition accuracy than the related work.",
  "keywords": [
    "Human activity recognition (HAR)",
    "localization",
    "long short-term memory (LSTM)",
    "people tracking",
    "temporal convolutional network (TCN)",
    "trajectory classification"
  ],
  "paper_institutions": [
    "Queen Mary University of London"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:47:03.431108Z",
  "template_tag": "simple",
  "summary": "### 摘要\n本文针对室内多人活动识别问题，提出使用2-D LIDAR采集人体运动轨迹并同时识别多个人的活动。该问题源于无线感知技术需区分人体与非人体目标并实现多目标跟踪的挑战。相关工作多依赖摄像头（受光照和隐私限制）、可穿戴设备（不便）或Wi-Fi/雷达（抗干扰差），而LIDAR具有厘米级精度和快速响应，适合室内多目标跟踪。方法上，首先对LIDAR原始点云使用DBSCAN聚类，提取15个几何特征并用随机森林区分人与非人；然后利用卡尔曼滤波器对多人连续跟踪获得轨迹；采用滑动窗口（窗口长度25点、间隔10点，对应2.5秒）分割轨迹，并通过空间变换（平移、旋转、缩放）和添加高斯噪声进行数据增强以解决类别不平衡。构建了LSTM网络（两层各120单元，全连接层200单元）和TCN（三个残差块，卷积核大小5，膨胀系数1、2、4，滤波器数75、150、300）用于将轨迹分类为15种厨房活动。实验在校园厨房进行，连续监测40小时，最多4人同时活动。增强后每类样本1000个，共15000个。训练集80%，测试集20%，并留出15%验证。TCN取得99.49%的总体准确率、99.53%召回率和99.51% F1值，略优于LSTM（99.39%准确率）；两者远优于SVM（95.88%）、HMM（85%）和DTW（90.9%）。轨迹增强使TCN和LSTM准确率分别提升约1.5%和1.7%。未来方向包括多LIDAR解决视线遮挡、融合可穿戴或雷达识别微观活动，以及探索少样本半监督学习和隐私保护。核心机制是TCN的膨胀因果卷积，公式为F(s)=(x*_d f)(s)=∑_{i=0}^{k-1} f(i)·x_{s-d·i}，通过增大膨胀因子扩大感受野。\n### 关键要点\n- 使用2-D LIDAR同时识别多人活动，通过聚类、人类识别、卡尔曼跟踪获得轨迹。\n- 采用DBSCAN对原始LIDAR点云聚类，无需预定簇数，可发现任意形状簇且抗噪。\n- 从每个簇提取15个几何特征，用随机森林（25棵树）区分人类与非人类，并利用置信度消除噪声。\n- 卡尔曼滤波实现多目标跟踪，状态向量包含位置、速度和加速度，预测与校正递推计算。\n- 轨迹分割使用滑动窗口，窗口长25点（2.5秒），间隔10点，兼顾实时性和信息量。\n- 数据增强采用平移、旋转、缩放和加高斯噪声，控制变换范围以保持语义，使每类样本增至1000个。\n- TCN由3个残差块组成，卷积核5，膨胀系数1、2、4，滤波器数75、150、300。\n- LSTM网络含2层（各120单元）和1个全连接层（200单元），输入形状为(25,2)。\n- 测试中TCN总体准确率99.49%，LSTM为99.39%，TCN收敛更快且损失曲线更平滑。\n- 增强后TCN和LSTM准确率分别提升约1.5%和1.7%，验证了数据增强的有效性。\n- 对比传统方法，TCN和LSTM的准确率、召回率和F1值均显著高于SVM、HMM和DTW。\n- 轨迹起点/终点关联四个停留区域（水槽、烤箱、椅子、门）以自动标注活动，但“进入”和“离开”需手动记录。"
}