Pedestrian re-identification method based on transfer learning and deep feature fusion

Abstract

The invention provides a pedestrian re-identification method based on transfer learning and deep feature fusion. The pedestrian re-identification method comprises the following steps of pre-training,human body posture correction and segmentation, feature vector, deep feature fusion, model training, model testing and result identification. According to the invention, global and local features of pedestrians are extracted by using a deep convolutional neural network; deep fusion is carried out on the two features to obtain final pedestrian feature representation; then, in the training process of the deep convolutional neural network, a pedestrian re-identification network model with a better effect is obtained by adopting a transfer learning mode, and finally, features extracted by the pedestrian re-identification network model have stronger resolution capability, so that the purpose of improving the pedestrian re-identification accuracy is achieved.

Description

Technical field [0001] The invention relates to a pedestrian re-identification method based on migration learning and deep feature fusion, and belongs to the technical field of deep learning and migration learning. Background technique [0002] Pedestrian re-recognition, the main purpose of the pedestrian matching task under the non-overlapping view domain multi-camera network, is to find out the target pedestrian captured by cameras in different positions at different times. [0003] With the development of artificial intelligence technology, pedestrian re-recognition technology in public security, image retrieval and other application scenarios has been widely concerned by research fields. However, compared with traditional biometric technologies such as face recognition and gesture recognition, pedestrian re-recognition technology is faced with factors such as low image resolution, viewing angle changes, posture changes, light changes, and occlusion due to the complex and uncont...

AI Technical Summary

Problems solved by technology

However, it is found in the research that most of the above pedestrian re-identification methods extract the global feature vector based on the global image of the pedestrian during the training process. At the same time, although some methods extract local features, they do not make full use of the local features of pedestrians for deep fusion. Discriminative Image Representation

Moreover, simple fine-tuning on the pedestrian database using the pre-trained model does not take into account the difference in data distribution between the source domain and the target domain dataset.

And then to the network migration effect is not ideal

Images