Person Re-Identification Using Color Features and CNN Features

Pedestrian re-identification task is a most challenge problem because the image angle, brightness and other factors change too much due to the different installation positions or resolutions of the camera. The traditional re-identification method, such as the manually annotating features, are not particularly desirable. In this paper, based on the idea of deep learning and feature representation, a neural network pedestrian re-identification algorithm combining color features and convolutional neural network extraction feature based on ResNet-50 model is proposed. Tested on a public pedestrian re-identification data set, experiments show that our method has a great improvement compared to the traditional pedestrian re-identification algorithm.