Pedestrian re-identification method and device based on residual attention mechanism space-time joint model

Abstract

The invention provides a pedestrian re-identification method and system based on a residual attention mechanism space-time joint model. The pedestrian re-identification method comprises the steps of performing feature extraction on an input pedestrian through a pre-trained ResNet-50 model; constructing a residual attention mechanism network, wherein a residual attention mechanism network comprisesa residual attention mechanism module, a feature sampling layer, a global pooling layer and a local feature connection layer; according to the trained residual attention mechanism network, using cosine distance to calculate a characteristic distance and recording the characteristic distance as a visual probability; performing modeling according to a camera ID and the frame number information space-time probability in the training sample pedestrian label, and performing Laplace smoothing processing on a probability model to acquire a pedestrian re-identification result by utilizing the visualprobability, the space-time probability and the final joint space-time probability. The pedestrian re-identification method based on the Bayesian joint probability solves the problem that an existingmethod ignores time and space prior information in a camera network, optimizes network iteration by using an attention residual mechanism network, accelerates convergence, and enables pedestrian re-identification precision to be higher through optimized Bayesian joint probability solution.

Description

technical field [0001] The invention belongs to the technical field of pedestrian re-identification, and relates to a pedestrian re-identification method and device based on a residual attention mechanism spatio-temporal joint model. Background technique [0002] Surveillance video is usually unable to obtain very high-quality face pictures due to the camera resolution and shooting angle, and pedestrians often show multi-scale characteristics, making detection and recognition difficult. At this time, pedestrian re-identification has become a very important alternative technology . Pedestrian re-identification is to achieve cross-device retrieval through a given monitored pedestrian image, which can effectively compensate for the visual limitations of fixed-view cameras, and has important application value in the fields of video surveillance, intelligent security, and smart cities. [0003] In recent years, with the development of machine learning theory, methods based on de...

AI Technical Summary

Problems solved by technology

However, most of the existing methods to solve the pedestrian re-identification problem only use the pedestrian identity information in the tag, and ignore the spatio-temporal information such as the camera ID information, the time sequence of the image, and the frame number of the image in the video, which are easier to collect.

Images