Research on Cross-modal Person re-ID Technology Based on Fusion of Multimodal Data

People recognition is a hot research direction of computer vision. To solve the problem that the existing heterogeneous modal migration fusion algorithm can not effectively compensate for the large semantic deviation between modes, a cross-modal Person re-ID algorithm based on multi-modal data fusion is proposed. Through the correlation matching of cross-modal features in each layer, the semantic deviation between heterogeneous modes is gradually reduced. The maximum correlation of top-level output features is used to optimize and adjust the modal network as a whole, which further improves the semantic relevance of modal depth. A cross-modal generation countermeasure network is designed, and end-to-end training is carried out, and the cross-modal discriminating infrared-visible pedestrian feature representation is learned to realize Person re-ID. Experimental results on data sets prove the effectiveness of this method.