A Pedestrian Re-Identification Method Based on Batch Block Occlusion Network
A pedestrian, network technology, applied in the field of computer vision, can solve problems such as increasing the complexity of the problem
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0045] A pedestrian re-identification method based on batch block occlusion network, comprising the following steps:
[0046] 1) Diversity: The images in the known pedestrian re-identification datasets Market-1501 and DukeMTMC-reID are divided into training datasets and test datasets. Market-1501 and DukeMTMC-reID are two large-scale pedestrian re-identification domains. General dataset, Market-1501 dataset contains 1501 identities observed from 6 camera viewpoints, contains 12936 training images detected by DPM of 751 persons and 19732 testing images of 750 persons; DukeMTMC-reID dataset There are 16,522 training images of 702 people and 17,661 test images of 702 people. They correspond to 1,404 different people, and the image sizes are different. Therefore, in this example, the training data set includes 29,458 images, and the test data set includes 37,393 images;
[0047] 2) Preprocessing: All images in the training data set and the test data set are cropped to a uniform si...
Embodiment 2
[0075] In step 2), k=32, and the remaining steps are the same as in Example 1.
[0076] The method of embodiment 1 is used below to compare the performance with existing methods, and the comparison results are as follows:
[0077] Table 1. Comparison of the effect of the method in this example and the existing pedestrian re-identification methods
[0078]
[0079] Table 2. Data comparison of global branch and feature deletion branch on Market-1501
[0080]
[0081] Among them, Baseline contains the global branch, BDB contains the global branch + Part 1 branch, and the method in this example contains the global branch + Part 1 branch + Part 2 branch.
[0082] It can be seen from the experimental results that the method in this example effectively improves the recognition accuracy of the network.
[0083] The experimental results of Example 1 and Example 2 are compared on Market-1501, such as Figure 4 As shown, when part=1, the learning feature of the occlusion module ...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


