Int J Performability Eng ›› 2020, Vol. 16 ›› Issue (7): 1027-1037.doi: 10.23940/ijpe.20.07.p5.10271037

Previous Articles     Next Articles

Pedestrian Re-Identification Incorporating Multi-Information Flow Deep Learning Model

Minghua Wei*   

  1. Department of Information and Technology, Fuzhou Polytechnic, Fuzhou, 350108, China
  • Submitted on ; Revised on ; Accepted on
  • Contact: * E-mail address:

Abstract: In the problem of pedestrian re-identification, it is difficult to extract effective pedestrian features and improve the re-identification accuracy due to changes in the angle of view, illumination, and pedestrian attitudes. However, the deep learning model is difficult to train and prone to the over-fitting problem when the training samples are few. To solve these problems, this paper proposes a multi-information flow convolutional neural network, the Mif-CNN) model, which contains a special convolutional structure. In this structure, the features extracted from each convolutional layer are connected to the input of all subsequent convolutional layers, which enhances the mobility of the network's feature information and the back propagation efficiency of the gradient and makes the pedestrian features extracted from the model more discriminative. The multi-loss function combination method is used to train the network model to distinguish the pedestrian categories better. Finally, the Euclidean distance is used to rank the pedestrian feature similarity. A number of experiments have been carried out on the pedestrian re-identification data sets i-LIDS and PRID-2011. The experimental results show that the algorithm proposed in this paper achieves the improvement of cumulative matching curve (CMC) and Rank-n re-identification rate in the comparison of images, videos, and deep learning models. Experimental results suggest that the proposed algorithm not only improves the accuracy of pedestrian re-identification in various scenes, but also enhances the ability of pedestrian features representation and effectively improves the over-fitting problem of the deep learning model.

Key words: deep neural networks, multi-information flow, pedestrian re-identification, multi-loss function combination, pedestrian features representation