CTS-YOLO: Real-Time Object Detection Algorithm Based on Improved YOLO11 in Complex Traffic Scenarios CTS-YOLO: Real-Time Object Detection Algorithm Based on Improved YOLO11 in Complex Traffic Scenarios

Traffic object detection in complex scenarios frequently requires adaptation to varying illumination and weather conditions, while also facing challenges such as small objects, occlusions, and blurring. These demands impose higher requirements on the precision and robustness of object detectors. To...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Signal, image and video processing Jg. 19; H. 10; S. 850
Hauptverfasser: Gao, Xingcheng, Huang, Keya, Li, Peiyu
Format: Journal Article
Sprache:Englisch
Veröffentlicht: London Springer London 01.10.2025
Springer Nature B.V
Schlagworte:
Accuracy
Algorithms
Computer Imaging
Computer Science
Datasets
Image Processing and Computer Vision
Modules
Multimedia Information Systems
Object recognition
Original Paper
Pattern Recognition and Graphics
Real time
Signal,Image and Speech Processing
Telematics
Traffic control
Vision
Weather
Object Detection
Autonomous Driving
YOLO11
Deep Learning
ISSN:1863-1703, 1863-1711
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Traffic object detection in complex scenarios frequently requires adaptation to varying illumination and weather conditions, while also facing challenges such as small objects, occlusions, and blurring. These demands impose higher requirements on the precision and robustness of object detectors. To address this, this paper improves the YOLO11n object detection algorithm and proposes the CTS-YOLO model. The main innovations and contributions of this algorithm include: the iRMA detection layer, which is a novel integration of EMA attention and an inverted residual structure specifically designed for small object detection; the SaEPPF module, a channel-aware multi-scale pooling mechanism; the SAC3k2 module, which enables dynamic multi-scale feature fusion; and the Inner-ShapeloU loss for shape-aware regression. Through the incorporation of these enhanced modules, the algorithm achieves a balance between improved accuracy and real-time performance. Experimental results demonstrate that CTS-YOLO outperforms baseline models on both the BDD100K and KITTI datasets, improving mAP@50 by 7.8% and 3.2%, respectively. Additionally, it achieves a real-time detection speed of 137 FPS on BDD100K, ensuring real-time performance.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1863-1703
1863-1711
DOI:10.1007/s11760-025-04490-0