A pedestrian re-identification method based on multi-channel attention features

Abstract

The invention discloses a pedestrian re-identification method based on multi-channel attention features, comprising the following steps: 1) constructing a convolutional neural network model based on channel attention, and pre-training a backbone network; 2) extracting pedestrian pictures in the backbone network 3) The weighted vector is multiplied by the output feature of the backbone network, and then added with it to obtain the channel attention feature; 4) Multiple attention features are repeatedly extracted, and the sea The Ringer distance is used to regularize the feature diversity; 5) The attention feature is input into the fully connected layer and the classifier, and the training minimizes the cross entropy loss and metric loss; 6) The test set image is input into the trained model to extract features, and the metric Sorting achieves pedestrian re-identification. The invention extracts the discriminative features of pedestrians based on the attention mechanism, limits the repeated extraction of similar attention features, and effectively improves the accuracy and robustness of pedestrian re-identification.

Description

technical field [0001] The invention relates to the technical fields of artificial intelligence and computer vision, in particular to a pedestrian re-identification method based on multi-channel attention features. Background technique [0002] With the acceleration of urbanization, the demand for social public security is increasing. Many important public places are covered by extensive camera networks. The use of computer vision technology to automate monitoring has become a hot spot of attention, and pedestrian re-identification technology has gradually become the focus of research. In general, given an image or video clip of a target pedestrian, person re-identification is used to retrieve the target pedestrian across the camera field of view. Due to the complexity and changeability of the monitoring scene, the collected pedestrian images often have difficulties such as illumination changes, perspective posture changes, and occlusions, which bring great challenges to ped...

AI Technical Summary

Problems solved by technology

Due to the inability of horizontal segmentation to accurately locate highly discriminative local parts, the accuracy performance is limited

In addition, the method using local part detection requires additional location labels, and the additional labeling work consumes manpower and material resources, which reduces the practicability of the method

Images