Siamese network-based method for pedestrian re-identification

Abstract

The invention discloses a Siamese network-based method for pedestrian re-identification. According to the method, two identical ResNet-50 convolutional neural networks are adopted to form a Siamese network, specifically, the ResNet-50 networks are pre-trained, the ResNet-50 networks are adjusted, the ResNet-50 networks are adopted to build the Siamese network, and the Siamese network is improved;a training dataset is pre-processed; the training dataset is adopted to train the Siamese network; an image to be checked is matched with images in the training dataset, so that an image pair can be obtained; pedestrian re-identification is carried out; and with the classification results of the two images and the consistency of the two images adopted as judgment conditions, images in the trainingdata set which belong to the same person as the image to be checked are determined. With the method of the invention adopted, the distinguishing degree of image depth features and the accuracy of pedestrian re-identification are improved, and time for querying images in a large-scale image set can be decreased.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a method for pedestrian re-identification based on a Siamese network. Background technique [0002] In recent years, with the popularization of video surveillance systems, video surveillance systems are playing an increasingly important role in combating crime and maintaining stability. Video investigation has become a new method for public security organs to investigate and handle cases. In video surveillance applications, the retrieval of specific suspects (especially people) is an important requirement. At present, this process is mainly completed manually, which consumes a lot of manpower, material resources and time, and affects the efficiency of investigation and case handling. The core key problem of specific target surveillance video retrieval - pedestrian re-identification refers to judging whether pedestrian images appearing under different surveillance cameras...

AI Technical Summary

Problems solved by technology

This processing method not only increases the workload of manual research and judgment, and reduces the automation of video analysis, but also, due to differences in viewing angles and lighting, the appearance characteristics of pedestrians will change greatly, which may cause the provided top results to be inconsistent. Reliability is lower

Images