Person Re-identification Based On Deep Multi-Feature Fusion Distance Learning

The purpose of person re-identification is to identify the same person captured in different positions and different camera views.However,person re-identification methods are essentially challenging because of the vulnerability of pedestrians to attitude,lighting,background and occlusion.In order to make the person re-identification method better,the main research contents of this paper include the following two aspects:(1)Based on the principle of deep multi-view feature distance fusion,a new pedestrian re-identification method is proposed.This method improves the recognition accuracy of pedestrian recognition by using the mutual cooperation of traditional features and depth features.Firstiy,the sliding frame technology is applied to the convolution layer,and the convolution features are processed at different scales to obtain a new low-dimensional depth region aggregate feature vector.The low-dimensional eigenvector with the dimension equal to the number of convolution layer channels is obtained.Secondly,from the perspective of the deep regional integration feature and the handcrafted feature,a distance learning algorithm was proposed by utilize the Cross-view Quadratic Discriminant Analysis metric learning.Finally,the weighted fusion strategy is used to accomplish the collaboration between handcrafted and deep convolution features.Through the comparative analysis of the experimental data,it is found that the accuracy of pedestrian re-recognition based on the distance-weighted fusion method is better than the recognition results based on a single feature distance metric,which further demonstrates the effectiveness of the depth region features and algorithm model proposed in this paper.(2)Using more complementary advantages of convolutional features at different levels,a pedestrian recognition method based on multi-scale convolutional feature fusion is proposed.In the training stage,the multi-layer convolution feature graph of the network is pooled by using different pooling strategies.At the same time,a series of optimization techniques and multiple loss functions are applied to improve the classification performance.Finally,the gradient descent method is used to optimize the obtained classification loss.In the test phase,the obtained features of different convolutional layers are stitched,and the fusion of multiple different feature vectors is used to predict the identity of the pedestrian.The proposed pedestrian recognition method based on multi-scale convolution feature fusion is tested on three large public pedestrian recognition datasets.The experimental results show that the multi-scale convolution feature extracted in this paper has a better recognition effect.