Occluded person re-identification based on foreground segmentation and multi-scale feature fusion

Re-identifying persons under occluded conditions is a demanding task within computer vision. This paper proposes a method called FGMS-Net, which significantly improves re-identification capability in occluded environments through several enhancements. First, an improved foreground segmentation technique is applied to effectively remove background and other clutter information, making feature extraction more precise. Second, to address the occlusion issue, a multi-scale feature discrimination approach is introduced, enabling the model to better capture local features and thus enhance recognition capability. Finally, an attention module is embedded within the backbone network., allowing it to focus more on critical information, which improves overall recognition performance. The proposed technique yields substantial performance gains in occluded person re-identification, as evidenced by experimental results showing a Rank-1 accuracy of 71.7% and an mAP of 61.6% on the Occluded-Duke MTMC dataset.