Person Re-identification based on Data Augmentation And Multiple Deep Feature Fusion

Person re-identification technology is taking a larger proportion in the research direction of pattern recognition. The main purpose of this technology is to identify the presence or absence of person in different time periods in different camera environments. At present, many research scholars have studied the feature representation of person in this technology, and sought and designed a person feature representation that is robust to factors such as viewing angle changes and illumination changes. The person feature extracted by many researchers have achieved very good results in the recognition. In order to better express the person feature, this paper proposes a Data Augmentation and Multiple Feature Fusion method (DAMFF). We consider that different deep convolution network have certain differences in the features extracted from the same person dataset. So we want to deeper fusion features of extracting from different deep convolution network that improve the recognition rate. Specifically, images increased by data augmentation for training and then the features are extracted by different deep convolution network, the final distance is obtained by a specific metric weighted fusion of different metric distance. Our method has achieved better results than other methods on datasets such as Market-1501, CUHK03.