EUVIMEDEUVIMED entdecken
Crossref

A Novel Evaluation Metric for Object Detection in Traffic Scenarios

Longjun Huang, Xueying Cai, Sheng Feng, Huxiong Li, Varshosaz Masood, Senang Ying, Binchao Zhou, Wentao Jia, Heng Liu, Lu Zhang, Yucheng Feng

IEEE Sensors Journal · 01.05.2026 · 26 (9) · 13322-13337

DOI: 10.1109/jsen.2026.3673625 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Sensors Journal, erschienen am 01.05.2026. Verfasst von Longjun Huang und 10 weiteren Autor:innen. Die Arbeit nennt 37 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Zhejiang Provincial Natural Science Foundation of China, Natural Science Foundation of Hebei Province.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate decision-making in autonomous driving relies on reliable object detection, particularly for small and edge-located targets such as roadside traffic signs, vehicles, and pedestrians. Although average precision (AP) is widely used to evaluate object detection performance, it overlooks object position, scale, and confidence distribution, providing only an overall measure of precision and recall. As a result, AP fails to adequately reflect detector performance in real-world traffic scenarios, especially for small-scale and edge objects. In this article, we proposed an improved metric called position-scale weighted AP (PSAP). Specifically, position and scale factors are incorporated into the definitions of true positives (TPs) and false positives (FPs), and the conventional precision–recall (PR) curve is extended to a 3-D formulation. This design enables a more fine-grained assessment of detection precision with respect to object location and scale. We evaluate PSAP on the PASCAL VOC dataset and a custom-labeled traffic dataset across several state-of-the-art object detectors. The experimental results indicate that RetinaNet exhibits comparatively weaker small object detection capability, leading to the largest discrepancy between AP (0.77) and PSAP (0.63). In contrast, YOLO-based detectors demonstrate a smaller gap between AP and PSAP. Furthermore, an independent comparative analysis of YOLO variants is conducted on the custom dataset, and statistical validation is performed to demonstrate the effectiveness of PSAP in providing additional insights into small and edge object detection quality.

Förderung

  • National Natural Science Foundation of China
  • Zhejiang Provincial Natural Science Foundation of China
  • Natural Science Foundation of Hebei Province
  • General Projects of Zhejiang Provincial Department of Education
  • Ministry of Education Theme Case Project in 2023

Zitieren

APA

Huang, L., Cai, X., Feng, S., Li, H., Masood, V., Ying, S., Zhou, B., Jia, W., Liu, H., Zhang, L., & Feng, Y. (2026). A Novel Evaluation Metric for Object Detection in Traffic Scenarios. IEEE Sensors Journal, 26(9), 13322-13337. https://doi.org/10.1109/jsen.2026.3673625

Vancouver

Huang L, Cai X, Feng S, Li H, Masood V, Ying S, et al. A Novel Evaluation Metric for Object Detection in Traffic Scenarios. IEEE Sensors Journal. 2026;26(9):13322-13337. doi:10.1109/jsen.2026.3673625

BibTeX
@article{Huang2026,
  author = {Huang, Longjun and Cai, Xueying and Feng, Sheng and Li, Huxiong and Masood, Varshosaz and Ying, Senang and Zhou, Binchao and Jia, Wentao and Liu, Heng and Zhang, Lu and Feng, Yucheng},
  title = {A Novel Evaluation Metric for Object Detection in Traffic Scenarios},
  journal = {IEEE Sensors Journal},
  year = {2026},
  volume = {26},
  number = {9},
  pages = {13322--13337},
  doi = {10.1109/jsen.2026.3673625}
}
Kostenlos starten →