Person Re-identification System by Selecting Appropriate Trained Model

Person Re-identification (ReID) involves verifying whether an individual captured across multiple cameras is the same person. While numerous studies focus on ReID, many assume environments with overlapping fields of view or uniform lighting conditions. However, security cameras in public facilities are often sparse in number and scattered due to cost and privacy considerations. This sparse arrangement results in varying angles of view and environmental contexts for each camera, posing challenges for re-identifying individuals using specific features. Consequently, ReID using specific features becomes difficult in such scenarios. To address this, our study proposes a system that performs feature extraction and ReID by dynamically selecting the optimal model for human images captured by multiple sparsely distributed cameras. By selecting a model suited to the shooting environment, we aim to improve ReID accuracy compared to using a single feature, thus enhancing the reliability of person identification across diverse camera setups.