| Popis: |
Person re-identification is the task to recognize the same person in gallery images from different cameras. Many previous researches aim to improve feature representation to separate different persons but features are only focus on local parts and a body part. Therefore, in this paper, we propose the Part-based Fusion Network (PFN) that extracted two global features from two layers of the ResNet50, split one global feature to form part-based features, and utilized both local and global features and concatenated to be a final feature for discriminating the same person. In addition, we combine the visual feature with the spatial temporal information to gain the better result on the testing phase. The experiment result shows that our method gained essential improvement and outperformed other state-of-the-art algorithms on two public datasets which are Market-1501 and DukeMTMC-reID. |
