Abstract: In the existing cross modal Re-ID (Person Re-identification) theories, few theories can use the attitude information in the images to go on the network learning. Considering the importance of attitude in Re-ID network learning, an end-to-end Re-ID theory is proposed which integrates local threshold binary image features. This method used ResNet50 as the backbone network to extract and fusion the features of three modal images. The network was trained by the cross entropy loss and improved hard triplet loss. The attitude information was used while using a simple network structure. The experimental results show that, the fusion of local threshold binary image information in the cross modal Re-ID network can enhance the accuracy of re-identification and improve the ability of finding the most difficult samples.