Smart Surveillance: Real-Time Shoplifting Detection Using Deep Learning and YOLOv8

Shoplifting is a major concern for retailers, leading to substantial financial losses worldwide. Traditional surveillance methods rely on manual monitoring, which is inefficient and prone to human error. This study presents an AI-powered shoplifting detection system using deep learning. The system integrates YOLOv8, accessed via the Roboflow API, for object detection, ByteTracker for object tracking, and EfficientNetV2B0 for behavioral classification. The DCSASS dataset from Kaggle was utilized for training, with a manually annotated subset of shoplifting and normal behavior videos ensuring balanced data. Preprocessing included converting videos into 224×224 frames, categorical encoding, and data augmentation techniques such as random flipping, rotation, zoom, and height & width scaling to enhance model generalization. The model achieved 93% accuracy, with a precision of 0.93, recall of 0.93, and F1-score of 0.93, demonstrating robust classification performance. Real-time deployment was achieved using Streamlit, enabling an interactive and efficient detection system. These results highlight the effectiveness of deep learning in automating shoplifting detection, offering a scalable, real-time solution for AI-driven retail security and loss prevention in modern retail environments.