Person re-identification from UAVs based on Deep hybrid features: Application for intelligent video surveillance

In the context of large-scale video surveillance systems, the person re-identification from UAVs is challenging due to the myriad appearance variation of a person across different UAVs. Actually, it is challenging issue to correctly match the same person in different UAV views. This paper introduced a new person re-identification method from UAVs that involves an offline phase and an inference phase. The first allows generating the person re-identification model, whereas the second aims to re-identify a person. Our method contributions are (1) the enhancement of the person re-identification robustness, using three complementary modules, namely the amplified occlusion, the hybrid discriminative features representation, and the bi-model deep metric learning; (2) the improvement of the proposed method effectiveness through relying on a new multi-shot-person retrieval module. Referring to the experimental assessment of the proposed method, the effectiveness and the efficiency of our method were evinced through a comparison with the state-of-the-art person re-identification methods.