Modern warehouse environments demand intelligent monitoring systems that ensure both operational safety and security. However, existing surveillance solutions remain largely reactive, relying on manual observation or isolated detection mechanisms that fail to address complex real-world challenges such as occluding theft behaviors, unsafe item placement, and varying lighting conditions. This research proposes an integrated vision-based framework that unifies warehouse safety monitoring and theft detection using advanced computer vision and deep learning techniques. The system combines object detection, human pose identification, human activity recognition (HAR), and multi-camera dynamic person re-identification in occlusion scenarios for theft detection and geometry-aware risk analysis and automated shelf edge detection to detect hazardous shelf conditions in real time. Theft-related activities such as loitering and abnormal human–item interactions are identified using activity sequences, while safety risks such as overhanging or fallen items are detected through segmentation-based object recognition and spatial boundary analysis. To enhance robustness, the framework incorporates multiple cameras for handling occluded situations and temporal stabilization techniques to reduce detection instability and false alerts. By integrating behavioral analysis with environmental risk assessment, the proposed system transforms traditional passive surveillance into a proactive monitoring solution. The framework aims to improve warehouse safety, reduce product damage, and enable early detection of theft through accurate, real-time alerts. This research contributes a scalable, multi-modal approach that addresses key limitations in existing systems, including lack of context awareness, poor occlusion handling, and absence of unified safety-security monitoring.