Pedestrian re-identification method based on multi-scale feature fusion
A multi-scale feature, pedestrian re-identification technology, applied in the field of pedestrian re-identification, can solve problems such as algorithm performance degradation, and achieve the effect of improving accuracy
Pending Publication Date: 2022-03-18
DALIAN UNIV OF TECH +2
View PDF0 Cites 2 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
However, pedestrian re-identification is a task that requires features rich in semantic information. Simple fusion of multi-scale features will lead to a decrease in algorithm performance. Therefore, it is necessary to find a fusion method that can not only guarantee semantic information, but also supplement detailed information. Improve the representation ability of the algorithm
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0035] In order to better explain the present disclosure and facilitate understanding, the present disclosure will be described in detail below through specific implementation manners in conjunction with the accompanying drawings. Apparently, the described embodiments are some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.
[0036] S1: Construct a deep learning network based on multi-scale fusion, and pre-train the multi-scale feature extractor ResNet50 on ImageNet. For the multi-scale feature X obtained by conv2, conv3, conv4, conv5 level output 1 ,X 2 ,X 3 ,X 4 . Set the size of the input image to 256×128×3, then the feature maps output at different levels are X 1 ∈ R 256×64×32 , X 2 ∈ R 512 ×32×16 , X 3 ∈ R 1024×16×8 , X4 ∈ R 2048×8×4 .
[0037] S2: Multi-s...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention discloses a pedestrian re-identification method based on multi-scale feature fusion, and the method comprises the steps: 1) constructing a model based on multi-scale fusion, and pre-training a backbone network and a multi-scale feature extractor; and 2) generating multi-scale features of the image by using a multi-scale feature extractor. And (3) fusing the features of two different scales by adopting a feature calibration model based on Transform. And 4) continuously fusing features of different levels from shallow features to deep features by using deep supervision fusion. And 5) supervising the fusion process by using cross entropy loss and triple loss. And 6) inputting a target test set image into the trained model to extract features, and sorting according to feature similarity to obtain a pedestrian re-identification result so as to realize pedestrian re-identification. According to the method, the convolutional neural network is used for extracting the multi-scale features, and the Transform is used for fusing the multi-scale information from a global angle, so that the features have detail and semantic information at the same time, and the accuracy of pedestrian re-identification is effectively improved.
Description
technical field [0001] The invention belongs to the field of image recognition, in particular to a pedestrian re-identification method for specific pedestrians in image retrieval. Background technique [0002] Person Re-identification (Person Re-identification) aims to retrieve the same pedestrian from reference images collected at different times from different camera perspectives. Pedestrian re-identification plays an important role in public security, intelligent video surveillance, and maintaining social security. Changes in human attributes and environmental factors such as posture, clothing, background, occlusion, and illumination lead to differences in the appearance of pedestrians under different cameras. [0003] With the development of deep learning technology, convolutional neural networks can achieve good performance in person re-identification tasks. However, due to the limited size of the convolution kernel, the convolutional neural network cannot extract the...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More IPC IPC(8): G06V20/52G06N3/04G06K9/62G06V10/774G06V10/80
CPCG06N3/045G06F18/214G06F18/253
Inventor 张国文戚金清王一帆卢湖川
Owner DALIAN UNIV OF TECH