Percepta: Adaptive Person Classification and Detection Under Real-World Adversarial Conditions

This record presents the research paper "PERCEPTA: Adaptive Person Classification and Detection Under Real-World Adversarial Conditions," which proposes a real-time person re-identification framework for intelligent surveillance. The framework integrates RT-DETR for person detection, a DINOv3 Vision Transformer optimized with Circle Loss for semantic feature learning, dynamic exemplar selection, and an FAISS-based semantic gallery for robust cross-camera identity management. Experimental evaluation demonstrates reliable performance under challenging conditions including motion blur, partial occlusion, illumination variation, and image noise. This work has been peer-reviewed and published in the Journal of Computer Science.