Pedestrian identification becomes difficult when facial cues are missing due to back-facing views, low light, or occlusions. This paper presents Deep-PPP, a pose-driven identification framework that relies solely on skeletal keypoints rather than facial or appearance features. Using OpenPifPaf for keypoint extraction, the system generates normalized pose-patched representations and embeds them into a discriminative pose-biometric descriptor integrating limb ratios, joint-angle invariants, and structural distances.
Evaluated across multiple benchmarks, Deep-PPP achieves 89.5% Rank-1 accuracy and 78.3% mAP on Market-1501, outperforming prior pose-guided methods such as PSE+GFL. Cross-dataset experiments further show a moderate generalization drop to CrowdPose due to increased occlusion, consistent with the challenging conditions of dense scenes. These results demonstrate that pose-based biometrics provide a practical, privacy-preserving alternative for pedestrian identification when facial information is unreliable or unavailable.