Pedestrian re-identification method based on feature fusion

Abstract

The invention provides a pedestrian re-identification method based on feature fusion. The pedestrian re-identification method based on feature fusion comprises the following steps: S1, preparing a pedestrian re-identification data set, and dividing the pedestrian re-identification data set into a training set tr, a verification value, a candidate pedestrian set g and a to-be-retrieved set q; s2, performing data enhancement on the data set; s3, according to the algorithm, a VGGNet-16 and DenseNet parallel fusion network is used as a pedestrian feature extraction network, and an ImageNet data set is used for training a preprocessing model; s4, extracting pedestrian features by using a parallel network; s5, classifying the image features obtained in the step 4 by using a Softmax classifier; and S6, the trained VGGNet-16 and DenseNet parallel fusion network is matched with a classifier model to carry out re-identification on the trained pedestrians. The pedestrian re-identification method based on feature fusion provided by the invention has the advantage that the acquisition efficiency and accuracy of the VGGNet-and DenseNet parallel fusion network are improved through the transmission mechanism and the dust removal mechanism.

Description

technical field [0001] The invention relates to the technical field of pedestrian re-identification based on feature fusion, in particular to a pedestrian re-identification method based on feature fusion. Background technique [0002] The purpose of the method based on feature representation is to extract robust features to represent pedestrians. The features used in pedestrian re-identification can be divided into three categories: visual features, filter features, and attribute features. The method based on metric learning is to learn two The similarity between pictures, the application of pedestrian re-identification is usually based on the feature representation, using the similarity between features to discriminate the similarity between pedestrian images, by learning a strong discriminative power The distance metric function is to make the distance between the same pedestrian as small as possible, and the distance between different pedestrians as large as possible. Fea...

AI Technical Summary

Problems solved by technology

[0004] Pedestrian re-identification technology has penetrated into many fields, such as pedestrian tracking, smart transportation, etc., but the current cameras are generally fixed on the equipment, such as the monitoring door. The width of the detection door is relatively large. Detection is carried out at an angle, and the camera will accumulate dust on the camera head during long-term use. Because of these factors, the effect of the pedestrian re-identification algorithm in practical applications is not satisfactory.

Images