DAN-YOLO: A Lightweight and Accurate Object Detector Using Dilated Aggregation Network for Autonomous Driving
01 natural sciences
0105 earth and related environmental sciences
DOI:
10.3390/electronics13173410
Publication Date:
2024-08-27T13:26:51Z
AUTHORS (7)
ABSTRACT
Object detection is becoming increasingly critical in autonomous driving. However, the accuracy and effectiveness of object detectors are often constrained by obscuration features details adverse weather conditions. Therefore, this paper presented DAN-YOLO vehicle detector specifically designed for driving conditions weather. Building on YOLOv7-Tiny network, SPP was replaced with SPPF, resulting SPPFCSPC structure, which enhances processing speed. The concept Hybrid Dilated Convolution (HDC) also introduced to improve ELAN-T structures, expanding network’s receptive field (RF) while maintaining a lightweight design. Furthermore, an efficient multi-scale attention (EMA) mechanism enhance feature fusion. Finally, Wise-IoUv1 loss function employed as replacement CIoU localization bounding box (bbox) convergence speed model. With input size 640 × 640, algorithm proposed study achieved increase mAP0.5 values 3.4% 6.3% compared BDD100K DAWN benchmark tests, respectively, achieving real-time (142.86 FPS). When other state-of-the-art detectors, it reports better trade-off terms under conditions, indicating suitability applications.
SUPPLEMENTAL MATERIAL
Coming soon ....
REFERENCES (41)
CITATIONS (0)
EXTERNAL LINKS
PlumX Metrics
RECOMMENDATIONS
FAIR ASSESSMENT
Coming soon ....
JUPYTER LAB
Coming soon ....