REMW‐YOLO: A Balanced Dense Pedestrian Detection Model With Efficient Feature Extraction and Fusion.
Saved in:
| Title: | REMW‐YOLO: A Balanced Dense Pedestrian Detection Model With Efficient Feature Extraction and Fusion. |
|---|---|
| Authors: | Yu, Hefeng1,2 (AUTHOR), Zu, Linan3 (AUTHOR), Pan, Xiaona1 (AUTHOR), Tang, Chao4 (AUTHOR), Jin, Ming Liang2 (AUTHOR) jinmingliang@qdu.edu.cn, Meng, Pingping1 (AUTHOR) mpp1979@qdu.edu.cn, Mohanta, Kallol (AUTHOR) kmohanta@wiley.com |
| Source: | International Journal of Intelligent Systems. 5/7/2026, Vol. 2026, p1-20. 20p. |
| Subjects: | Feature extraction, Data fusion (Statistics), Object recognition (Computer vision), Loss functions (Statistics) |
| Abstract: | Dense pedestrian detection plays a crucial role in intelligent security and video analysis, yet it faces challenges such as occlusion, scale variations, and difficulties in feature extraction. To address these, this study proposes REMW‐YOLO, an efficient model built on YOLO11n that balances performance and computational efficiency. To tackle occlusion, we design the C3k2_RD module, replacing the traditional C3k2 structure. This module integrates RFAConv and DynamicConv to form a complementary "local enhancement–global adaptation" mechanism, improving detection of occluded pedestrians. We also present the efficient dual‐channel fusion feature pyramid network (EDFFPN), a lightweight architecture using dual‐channel fusion mechanisms to preserve detailed features while addressing multiscale detection and reducing computational cost. Additionally, we develop the multiscale lightweight decoupled detection head (MSLDH), which enhances feature expression for dense targets without increasing complexity. Finally, we propose the Wise‐Inner‐MPDIoU loss function to address sample imbalance in dense scenes, resulting in faster convergence and more precise localization. Experiments on the WiderPerson, CrowdHuman, and CityPersons datasets show that REMW‐YOLO improves mAP@0.5 by 2.11%, 1.81%, and 2.90%, and mAP@0.5:0.95 by 1.81%, 2.36%, and 2.19%, respectively, over YOLO11n, with overall performance significantly surpassing other methods. Cross‐dataset validation confirms the model's generalization. REMW‐YOLO offers a balanced, efficient solution for dense pedestrian detection. [ABSTRACT FROM AUTHOR] |
| Copyright of International Journal of Intelligent Systems is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | Engineering Source |
|
Full text is not displayed to guests.
Login for full access.
|
|
| Abstract: | Dense pedestrian detection plays a crucial role in intelligent security and video analysis, yet it faces challenges such as occlusion, scale variations, and difficulties in feature extraction. To address these, this study proposes REMW‐YOLO, an efficient model built on YOLO11n that balances performance and computational efficiency. To tackle occlusion, we design the C3k2_RD module, replacing the traditional C3k2 structure. This module integrates RFAConv and DynamicConv to form a complementary "local enhancement–global adaptation" mechanism, improving detection of occluded pedestrians. We also present the efficient dual‐channel fusion feature pyramid network (EDFFPN), a lightweight architecture using dual‐channel fusion mechanisms to preserve detailed features while addressing multiscale detection and reducing computational cost. Additionally, we develop the multiscale lightweight decoupled detection head (MSLDH), which enhances feature expression for dense targets without increasing complexity. Finally, we propose the Wise‐Inner‐MPDIoU loss function to address sample imbalance in dense scenes, resulting in faster convergence and more precise localization. Experiments on the WiderPerson, CrowdHuman, and CityPersons datasets show that REMW‐YOLO improves mAP@0.5 by 2.11%, 1.81%, and 2.90%, and mAP@0.5:0.95 by 1.81%, 2.36%, and 2.19%, respectively, over YOLO11n, with overall performance significantly surpassing other methods. Cross‐dataset validation confirms the model's generalization. REMW‐YOLO offers a balanced, efficient solution for dense pedestrian detection. [ABSTRACT FROM AUTHOR] |
|---|---|
| ISSN: | 08848173 |
| DOI: | 10.1155/int/2375277 |
