AI-Based Customer Behavior Monitoring and Suspicious Activity Detection in Retail Environments

DOI

https://www.doi.org/10.59256/ijsreat.20260606021

Published In

Volume 6, Issue 6 (2026)

Published Online

November-December 2026

Pages

153-164

Article Type

Original Article

Check for updates

Retail environments use CCTV cameras to observe customer activity, yet continuous manual monitoring of several video feeds can be demanding and may lead to missed events. This paper presents a proposed AI-based customer behavior monitoring framework for retail stores, supermarkets, and shopping malls. The design combines YOLO-based object detection, multi-object tracking using DeepSORT or ByteTrack, human pose estimation using MediaPipe Pose or YOLO-Pose, and temporal action recognition using ST-GCN or SlowFast. Video frames are acquired from CCTV, webcam, or IP-camera sources and processed to detect people and selected products. Tracking assigns temporary identifiers to visible people, while pose estimation extracts body landmarks such as wrists, elbows, and shoulders. A temporal recognition stage is intended to classify activities represented in its training data, including reaching toward a pocket or potentially concealing an item. Predictions exceeding a configured confidence threshold are forwarded to an alert dashboard with timestamps and relevant evidence for human review. The supplied project report describes this as a proposed system; it does not provide measured accuracy, benchmark results, or a verified test dataset. Accordingly, this manuscript documents the design, workflow, evaluation methodology, and limitations without claiming experimentally validated performance. Automated alerts are treated as review cues and not as proof of theft.