Pedestrian re-identification feature extraction method based on multi-scale feature fusion

Abstract

The invention relates to feature extraction in the field of pedestrian re-identification, in particular to a pedestrian re-identification feature extraction method based on multi-scale feature fusion.The method of the present invention comprises the following steps of improving and constructing a multi-scale feature fusion network by taking ResNet as a backbone, wherein features extracted by a layer 3 are used as shared features, two branches are followed, the two branches are respectively a Global branch and a Part branch; reducing the stride of the last layer of ResNet from 2 to 1 by the Part branch; operating the Globe branch by using a layer 4 of the original ResNet; extracting feature maps of the layer3.1 and the layer4.1, and respectively fusing a Part branch and a Global branch with feature maps of a layer 3.1 and a layer 4.1; respectively marking as layer3.1 _ p and layer4.1 _ g, and marking as layer3.1 _ p and layer4.1 _ g; subjecting the four feature vectors (layer4.4 (Part), layer4.4 (Global), layer3.1 _ p and layer4.1 _ g) to dimension reduction to form 512-dimensional features for feature fusion, and forming 2048-dimensional features for similarity measurement. According to the method, the underlying feature map can be utilized to contain more tiny detail information of the pedestrian image, so that the extracted features can distinguish similar pedestrians more easily, and the robustness of the extracted features is enhanced.

Description

technical field [0001] The invention relates to the fields of computer vision technology and pedestrian re-identification. In particular, it involves a pedestrian feature extraction method based on multi-scale feature fusion. Background technique [0002] It has become a consensus to use high-tech means to strengthen social management and prevent crimes from happening. To this end, governments around the world have installed a large number of cameras at key points in public places, traffic intersections, living quarters, parking lots, etc. to strengthen the observation and identification of pedestrian behavior. Cameras generate huge amounts of data every day, and it is of great significance to analyze these data. However, it is particularly difficult to obtain biological characteristics such as face and gait in complex scenes, and pedestrian re-identification technology has emerged for this reason. Different from face image recognition technology in the traditional sense,...

AI Technical Summary

Problems solved by technology

[0007] In view of the fact that the features extracted by existing algorithms in the field of pedestrian re-identification are still not robust and the generalization ability of the model is weak, the present invention extracts robust pedestrian representations by integrating multi-scale features. The algorithm architecture is as follows: figure 2 As shown, the application value of pedestrian re-identification technology in intelligent security and intelligent business can be improved

Images