Pedestrian re-identification method based on video appearance and motion information synchronous enhancement

Abstract

The invention discloses a pedestrian re-identification method based on video appearance and motion information synchronous enhancement. During training, pedestrian appearance and motion information ina backbone network are respectively enhanced through an appearance enhancement module AEM and a motion enhancement module MEM. The appearance enhancement module AEM uses an attribute recognition model obtained by training an existing large-scale pedestrian attribute data set to provide an attribute pseudo label for the large-scale pedestrian video data set, and enhances appearance and semantic information through attribute learning; the motion enhancement module MEM predicts pedestrian gait information by using a video prediction model, enhances gait information features with identity discrimination ability in a pedestrian feature extraction backbone network, and improves pedestrian re-identification performance. In practical application, only the pedestrian feature extraction backbone network needs to be reserved, and higher pedestrian re-identification performance can be obtained without increasing the network complexity and the model size. And the enhanced backbone network featuresobtain higher accuracy in a video-based pedestrian re-identification task.

Description

technical field [0001] The invention belongs to the technical field of intelligent identification, in particular to a pedestrian re-identification method based on synchronous enhancement of video appearance and motion information. In the part of algorithm design and model training, deep learning technology is involved. Background technique [0002] The task of pedestrian re-identification is to retrieve the pedestrian target in camera B where there is no overlapping area of ​​camera A, and find the pedestrian target appearing in camera A again. As an important research direction and research hotspot at present, pedestrian re-identification has a wide range of applications in the fields of intelligent monitoring, smart city, public security prevention and criminal investigation, such as cross-camera pedestrian tracking and behavior analysis, and image retrieval of suspects or interested persons. with queries etc. [0003] With the rapid development of deep learning, more an...

AI Technical Summary

Problems solved by technology

This method adopts a deep neural network model, uses attribute learning and gait prediction to enhance the appearance information and motion information in the pedestrian feature extraction backbone network, and solves the problem of insufficient utilization of video data information by the pedestrian feature extraction backbone network in the existing methods problems, and fully improve the performance of backbone network feature extraction and pedestrian re-identification

Images