PoseLift
收藏资源简介:
PoseLift数据集是由北卡罗来纳大学夏洛特分校与零售店合作创建的隐私保护数据集,专注于商店盗窃行为的检测。该数据集包含155个视频,涵盖了真实零售环境中的正常购物和盗窃行为,视频分辨率为1920×1080,帧率为15帧/秒。数据集通过隐私保护技术对数据进行匿名化处理,提供了姿态序列数据,包括人体姿态、边界框和跟踪ID等信息。PoseLift的创建过程包括从零售店的CCTV录像中提取数据,并通过人工标注对每帧进行分类,标注为正常购物或盗窃行为。该数据集旨在通过基于姿态的异常检测方法解决零售安全中的商店盗窃问题,为计算机视觉研究提供了一个重要的工具。
The PoseLift Dataset is a privacy-preserving dataset jointly created by the University of North Carolina at Charlotte and retail stores, focusing on shoplifting detection. It contains 155 videos covering both normal shopping and shoplifting behaviors in real retail environments, with a resolution of 1920×1080 and a frame rate of 15 frames per second. The dataset has been anonymized via privacy-preserving technologies, and provides pose sequence data including human poses, bounding boxes, tracking IDs and other relevant information. The development process of PoseLift includes extracting data from CCTV footage of retail stores and manually annotating each frame to classify it as either normal shopping or shoplifting. This dataset aims to address the shoplifting problem in retail security through pose-based anomaly detection methods, serving as a critical tool for computer vision research.

- 1Exploring Pose-Based Anomaly Detection for Retail Security: A Real-World Shoplifting Dataset and Benchmark北卡罗来纳大学夏洛特分校 · 2025年



