Advancing Person Re-Identification through Pose-Estimated Features

This work presents a novel framework for person re-identification that leverages 2D pose estimation to enhance the accuracy and robustness of the existing method. By incorporating pose estimation, the proposed method captures the spatial relationships between object key points, enabling a more comprehensive understanding of object configurations. We employ a CNN-based model, using a bounding box as the backbone for the detector. The model’s ability to generate precise person class facilitates accurate pose estimation by establishing associations between key points and then using those features to identify the same person from different camera angles. As the technology of computer vision develops, employing pose estimation in identifying a person has tremendous potential for improving the capabilities of a wide range of real-world applications.