Shoplifting represents a persistent and costly issue within the retail industry, contributing to considerable financial losses annually. To address this problem, this research introduces an intelligent surveillance system that integrates real-time object detection, human pose estimation, and behavioral analysis to proactively identify potential shoplifting incidents. The proposed system utilizes the YOLOv5 deep learning model for detecting individuals and store items in video feeds, MediaPipe for precise body pose landmark tracking, and a Random Forest classifier to categorize human behaviors as either suspicious or normal. The experimental results indicate that the system achieves a high detection accuracy with a minimal false positive rate, demonstrating its practical effectiveness in real-world retail scenarios.