Pedestrian re-identification method based on global features and local features of attention mechanism

Abstract

The invention relates to a pedestrian re-identification method based on global features and local features of an attention mechanism. The pedestrian re-identification method comprises the steps of respectively extracting the global features and the local features of pedestrians; in the global feature branch, taking the whole pedestrian feature image as input, sending the pedestrian feature image into a space attention mechanism module and a channel attention mechanism module, and fusing the feature representations of the two modules; in the local feature branch, horizontally and averagely dividing the pedestrian feature map into three parts, and inputting the three divided parts into a channel attention mechanism module to obtain the local feature of each part; sending the global feature and the local feature into a feature vector extraction module to obtain a feature vector for pedestrian prediction; and training the whole network to obtain a pedestrian re-identification model. According to the method, the global features and the local features of the pedestrian images are fully utilized, the attention mechanism is effectively fused, the pedestrian features have better discrimination ability, a good pedestrian re-identification result is obtained, and the model matching accuracy is improved.

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 global features and local features of an attention mechanism. Background technique [0002] Pedestrian re-identification technology refers to the technology of judging whether the pedestrian images or videos captured by different cameras are the same pedestrian through computer vision technology. Because of its significance in video surveillance and security reconnaissance, pedestrian re-identification technology has attracted widespread attention and is the key to building a harmonious society and a safe city. [0003] Traditional pedestrian re-identification technology can be divided into two main points: one is feature extraction, which extracts discriminative features from pedestrian images or videos captured by cameras; the other is distance metric learning, which makes the same pedestrian fe...

AI Technical Summary

Problems solved by technology

[0004] Most of the current pedestrian re-identification tasks are based on the global features of pedestrian images, and the background of pedestrian images is very complex, some inconspicuous details are easy to be ignored, and the diversity of pedestrian re-identification datasets is small, which limits pedestrian recognition. re-identification accuracy

However, some pedestrian re-identification methods based on local feature extraction can obtain robust results in the face of partial changes and occlusions, but they ignore the global context dependence in the pedestrian re-identification image and reduce the matching accuracy.

Images