YOLO-RTLite: A TensorRT-Optimized Lightweight Detector for Real-Time Road Object Detection
International Journal of Computing · 30.06.2026 · 362-368
DOI: 10.47839/ijc.25.2.4662 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Computing, online erschienen am 30.06.2026. Verfasst von Manjot Kaur und 2 weiteren Autor:innen.
Abstract (Originalsprache)
This paper presents YOLO-RTLite, an optimized lightweight object detection model based on YOLOv4-Tiny. The proposed model addresses the challenges of accurate and reliable object detection in curved and blind road conditions, where visibility is limited and timely detection is critical. YOLO-RTLite is specifically engineered for real-time identification of critical road obstacles, focusing on vehicles and animals, to reduce road accidents and improving transportation safety. The YOLO-RTLite architecture incorporates an additional convolutional layer (with Mish activation function) and three novel skip connections to reinforce multi-scale feature propagation and feature reuse. Model performance is further improved through selective tuning of filter sizes and anchor boxes which enable better adaptation to diverse object dimensions. For real-time deployment, TensorRT-based acceleration is utilized to significantly increase inference speed. Performance results demonstrate that YOLO-RTLite achieves high precision (~95%), accuracy (~99%), robust recall (~98%), and efficient inference time (0.165 s) which is a significant improvement over its baseline model (YOLOv4 tiny). These performance stats support its suitability for safety-critical intelligent transportation systems.
Zitieren
APA
Kaur, M., Randhawa, R., & Kaur, J. (2026). YOLO-RTLite: A TensorRT-Optimized Lightweight Detector for Real-Time Road Object Detection. International Journal of Computing, 362-368. https://doi.org/10.47839/ijc.25.2.4662
Vancouver
Kaur M, Randhawa R, Kaur J. YOLO-RTLite: A TensorRT-Optimized Lightweight Detector for Real-Time Road Object Detection. International Journal of Computing. 2026:362-368. doi:10.47839/ijc.25.2.4662
BibTeX
@article{Kaur2026,
author = {Kaur, Manjot and Randhawa, Rajneesh and Kaur, Jagroop},
title = {YOLO-RTLite: A TensorRT-Optimized Lightweight Detector for Real-Time Road Object Detection},
journal = {International Journal of Computing},
year = {2026},
pages = {362--368},
doi = {10.47839/ijc.25.2.4662}
}
EUVIMED