Search Results (3)

Video-based pedestrian re-identification (Re-ID) is used to re-identify the same person across different camera views. One of the key problems is to learn an effective representation for the pedestrian from video. However, it is difficult to learn an effective representation from one single modality of a feature due to complicated issues with video, such as background, occlusion, and blurred scenes. Therefore, there are some studies on fusing multimodal features for video-based pedestrian Re-ID. However, most of these works fuse features at the global level, which is not effective in reflecting fine-grained and complementary information. Therefore, the improvement in performance is limited. To obtain a more effective representation, we propose to learn fine-grained features from different modalities of the video, and then they are aligned and fused at the fine-grained level to capture rich semantic information. As a result, a multimodal token-learning and alignment model (MTLA) is proposed to re-identify pedestrians across camera videos. An MTLA consists of three modules, i.e., a multimodal feature encoder, token-based cross-modal alignment, and correlation-aware fusion. Firstly, the multimodal feature encoder is used to extract the multimodal features from the visual appearance and gait information views, and then fine-grained tokens are learned and denoised from these features. Then, the token-based cross-modal alignment module is used to align the multimodal features at the token level to capture fine-grained semantic information. Finally, the correlation-aware fusion module is used to fuse the multimodal token features by learning the inter- and intra-modal correlation, in which the features refine each other and a unified representation is obtained for pedestrian Re-ID. To evaluate the performance of fine-grained features alignment and fusion, we conduct extensive experiments on three benchmark datasets. Compared with the state-of-art approaches, all the evaluation metrices of mAP and Rank-K are improved by more than 0.4 percentage points. Full article

Recently, person-following robots have been increasingly used in many real-world applications, and they require robust and accurate person identification for tracking. Recent works proposed to use re-identification metrics for identification of the target person; however, these metrics suffer due to poor generalization, and due to impostors in nonlinear multi-modal world. This work learns a domain generic person re-identification to resolve real-world challenges and to identify the target person undergoing appearance changes when moving across different indoor and outdoor environments or domains. Our generic metric takes advantage of novel attention mechanism to learn deep cross-representations to address pose, viewpoint, and illumination variations, as well as jointly tackling impostors and style variations the target person randomly undergoes in various indoor and outdoor domains; thus, our generic metric attains higher recognition accuracy of target person identification in complex multi-modal open-set world, and attains 80.73% and 64.44% $Rank$-1 identification in multi-modal close-set PRID and VIPeR domains, respectively. Full article

Person re-identification is the task of recognizing a subject across different non-overlapping cameras across different views and times. Most state-of-the-art datasets and proposed solutions tend to address the problem of short-term re-identification. Those models can re-identify a person as long as they are wearing the same clothes. The work presented in this paper addresses the task of long-term re-identification. Therefore, the proposed model is trained on a dataset that incorporates clothes variation. This paper proposes a multi-modal person re-identification model. The first modality includes soft bio-metrics: hair, face, neck, shoulders, and part of the chest. The second modality is the remaining body figure that mainly focuses on clothes. The proposed model is composed of two separate neural networks, one for each modality. For the first modality, a two-stream Siamese network with pre-trained FaceNet as a feature extractor for the first modality is utilized. Part-based Convolutional Baseline classifier with a feature extractor network OSNet for the second modality. Experiments confirm that the proposed model can outperform several state-of-the-art models achieving 81.4 % accuracy on Rank-1, 82.3% accuracy on Rank-5, 83.1% accuracy on Rank-10, and 83.7% accuracy on Rank-20. Full article