Study on Pedestrian Detection Methods Based on Improved CBAM-YOLOv7

In view of the problem of YOLOv7 detection accuracy, a pedestrian detection model improving YOLOv7 network is proposed. First, the backbone network is downsampled using modules constructed with SPPCSPC module and CBAM to reduce the loss of fine-grained feature information. Secondly, the multiscale d...

Full description

Saved in:
Bibliographic Details
Published in2024 6th International Conference on Communications, Information System and Computer Engineering (CISCE) pp. 1244 - 1247
Main Authors Song, Yang, Zhang, Ruoyuan
Format Conference Proceeding
LanguageEnglish
Published IEEE 10.05.2024
Subjects
Accuracy
Cameras
CBAM
component
Computational modeling
loss function
Pedestrian detection
Pedestrians
Performance evaluation
Testing
Training
YOLOv7
Online AccessGet full text

Cover

More Information
Summary:In view of the problem of YOLOv7 detection accuracy, a pedestrian detection model improving YOLOv7 network is proposed. First, the backbone network is downsampled using modules constructed with SPPCSPC module and CBAM to reduce the loss of fine-grained feature information. Secondly, the multiscale detection capability of the model is enhanced by increasing the small-size detection layer. Then, the original CIoU loss function was replaced with the \mathbf{\alpha}-\mathbf{EIoU} loss function to improve the pedestrian target positioning accuracy. Using the Crowdhuman dataset for training and testing, the experimental results show that the proposed algorithm improves 5.2% and 4.3% recall and average accuracy values over the original algorithm, respectively, which can effectively improve the accuracy of pedestrian detection in long-distance targets and dense scenarios.
ISSN:2833-2423
DOI:10.1109/CISCE62493.2024.10653343