Cross-modal pedestrian re-identification method based on multi-modal common feature space exploration

The invention discloses a cross-modal pedestrian re-identification method based on multi-modal common feature space exploration, which comprises a single-modal feature extraction network, a multi-modal common feature space exploration network and a multi-modal feature interaction auxiliary training module, and is used for preprocessing a training image to obtain a training sample; a single-modal feature extraction network and a multi-modal common feature exploration network are constructed, the single-modal feature extraction network performs single-modal feature extraction, and the multi-modal common feature space exploration network extracts modal common features from multiple modal common feature spaces and performs reasonable fusion. The training of the two parts of networks is carried out under the supervision of classification loss and measurement loss, and the obtained feature description is used for carrying out similarity calculation when the test is carried out; according to the method, the modal di