YOLO series object detection networks optimized with luminance attention mechanism network

Jie Zeng; Kan Wang; Xiong Hu; Yuanzhi Hu; Xi Liu; Zhengqian Cheng

doi:10.1117/12.3016067

20 February 2024 YOLO series object detection networks optimized with luminance attention mechanism network

Jie Zeng, Kan Wang, Xiong Hu, Yuanzhi Hu, Xi Liu, Zhengqian Cheng

Proceedings Volume 13064, Seventh International Conference on Traffic Engineering and Transportation System (ICTETS 2023); 130642I (2024) https://doi.org/10.1117/12.3016067
Event: 7th International Conference on Traffic Engineering and Transportation System (ICTETS 2023), 2023, Dalian, China

Abstract

The variation in lighting conditions significantly affects the performance of deep learning-based visual object detection algorithms. Under strong light, weak light, and dark conditions, the algorithm's effectiveness tends to decrease. This paper proposes a neural network module based on luminance attention mechanism, which adaptively adjusts the network's attention weights for different regions of an image based on their luminance. It enhances the network's attention on areas with significant differences in luminance and background, thereby improving the robustness of the object detection algorithm to lighting changes. Experimental results on the DAIR-V2X vehicle-road cooperative dataset demonstrate that the average mAP@.5 of the YOLO series object detection network is improved by 4.54% with the addition of the proposed luminance attention mechanism module. Qualitative and quantitative analyses validate that the proposed algorithm module enhances the performance of YOLO series object detection networks in scenarios with lighting variations.

Citation Download Citation

Jie Zeng, Kan Wang, Xiong Hu, Yuanzhi Hu, Xi Liu, and Zhengqian Cheng "YOLO series object detection networks optimized with luminance attention mechanism network", Proc. SPIE 13064, Seventh International Conference on Traffic Engineering and Transportation System (ICTETS 2023), 130642I (20 February 2024); https://doi.org/10.1117/12.3016067

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available