Pedestrian Re-Identification Method Based on Enhanced Deep Convolutional Neural Network

Abstract

The invention relates to a pedestrian re-identification method based on an enhanced deep convolutional neural network, which uses a basic deep learning convolutional neural network model to extract the basic depth features of pedestrian images, and uses traditional manual feature extraction methods to extract manual features of pedestrian images and Dimensionality reduction; apply the feature reconstruction module to fuse the basic deep features and manual features into enhanced deep features; predict whether the pedestrians in the two images are the same person through feature comparison, and jointly use the classification loss function and verification loss function to classify the input image and Similarity-difference verification, which trains the network with the goal of minimizing the joint loss, so that the network generates more discriminative pedestrian image features. The invention makes full use of the complementarity between manual features and deep features, and proposes a strategy of jointly using classification loss and verification loss functions for supervised network training, which achieves good performance and effectively improves the accuracy of pedestrian re-identification.

Description

technical field [0001] The invention belongs to the technical field of computer vision pedestrian re-identification, in particular to a pedestrian re-identification method based on an enhanced deep convolutional neural network. Background technique [0002] With the needs of social security and the development of science and technology, public places such as airports, stations, shopping malls and schools have deployed a large number of camera networks. These cameras with a large geographical space span and non-overlapping monitoring areas provide a large amount of video data for the subsequent processing system. In this context, relying on manual processing of these data becomes inefficient and infeasible, and must rely on advanced machine algorithms for intelligent processing. Automatically analyzing these video data through machine algorithms can not only improve efficiency but also significantly improve the quality of monitoring. Pedestrian re-identification is an import...

AI Technical Summary

Problems solved by technology

At the same time, it is a challenging task because the visual appearance of pedestrians can vary significantly from camera to camera due to pedestrian pose changes, different camera perspectives, lighting differences, occlusions, and background disturbances.

Images