Exploring Pose-Based Anomaly Detection for Retail Security: A Real-World Shoplifting Dataset and Benchmark

TL;DR

This work reframes shoplifting detection as an anomaly-detection task and introduces PoseLift, a real-world, privacy-preserving dataset of anonymized pose sequences collected from a six-camera retail environment. PoseLift provides frame-level shoplifting annotations and pose data (17 keypoints per person) derived from HRNet, with bounding boxes from YOLOv8 and tracking from ByteTrack, enabling analysis while protecting user identities. The authors benchmark three state-of-the-art pose-based anomaly detectors (STG-NF, TSGAD, GEPC) on PoseLift, reporting AUC-ROC, AUC-PR, and EER metrics; STG-NF achieves the strongest performance (AUC-ROC 67.46%, AUC-PR 84.06%, EER 0.39), demonstrating the feasibility of privacy-conscious, real-world shoplifting detection. PoseLift stands as a valuable, publicly available resource to advance ethical computer vision in retail security and to spur development of robust, bias-resistant anomaly-detection approaches in multi-view environments.

Abstract

Shoplifting poses a significant challenge for retailers, resulting in billions of dollars in annual losses. Traditional security measures often fall short, highlighting the need for intelligent solutions capable of detecting shoplifting behaviors in real time. This paper frames shoplifting detection as an anomaly detection problem, focusing on the identification of deviations from typical shopping patterns. We introduce PoseLift, a privacy-preserving dataset specifically designed for shoplifting detection, addressing challenges such as data scarcity, privacy concerns, and model biases. PoseLift is built in collaboration with a retail store and contains anonymized human pose data from real-world scenarios. By preserving essential behavioral information while anonymizing identities, PoseLift balances privacy and utility. We benchmark state-of-the-art pose-based anomaly detection models on this dataset, evaluating performance using a comprehensive set of metrics. Our results demonstrate that pose-based approaches achieve high detection accuracy while effectively addressing privacy and bias concerns inherent in traditional methods. As one of the first datasets capturing real-world shoplifting behaviors, PoseLift offers researchers a valuable tool to advance computer vision ethically and will be publicly available to foster innovation and collaboration. The dataset is available at https://github.com/TeCSAR-UNCC/PoseLift.

Figures (7)

• Figure 1: Retail revenue lost to shoplifting in the USA (in billions), with losses reaching nearly $122 billion in 2023 and projected to surpass $143 billion by 2025 shoplifting_capitalone2024national_retail_federation.
• Figure 2: Bird's-eye view of the retail store, illustrating the locations of six cameras and their coverage area.
• Figure 3: Segmented images from six camera views within a retail store, showcasing various perspectives used to capture normal and shoplifting instances in the dataset.
• Figure 4: Four examples of normal shopping behavior, including browsing shelves (4.1), walking through the store (4.2), picking up an item (4.3), and carrying a bottle (4.4). These instances represent typical actions captured in PoseLift for comparison with shoplifting scenarios.
• Figure 5: Four shoplifting instances captured from different angles, including concealing an item in their pants, placing an item in a front pocket, putting an item in an open bag, and placing an item in a bag on the ground. These instances illustrate typical shoplifting behaviors featured in our dataset.