Evaluating Pretrained YOLO11s Vehicle Detection Under Daylight and Simulated Nighttime Conditions Using Lightweight Image Enhancement

Vehicle Detection, YOLO11s, Intelligent Transportation Systems, Advanced Driver Assistance Systems, Low-Light Vision, Image Enhancement, CLAHE, Gamma Correction.

Authors

  • Marwa K. Hasan Technical Engineering College - Kirkuk, Northern Technical University, Kirkuk, 36001, Iraq
  • Lana OLana O. Alhamawndi Technical Engineering College - Kirkuk, Northern Technical University, Kirkuk, 36001, Iraq
  • Maroa Essam Baker Technical Engineering College - Kirkuk, Northern Technical University, Kirkuk, 36001, Iraq
August 29, 2026
September 2, 2026

Downloads

Vehicle detection is an important perception task in Intelligent Transportation Systems (ITS) and Advanced Driver Assistance Systems (ADAS), and the performance under different illumination conditions still remains as an important challenge. The current deep-learning object detectors perform well under daylight conditions, while the detection behavior may be degraded in nighttime and low-light conditions due to low contrast and limited object visibility. This paper introduces a lightweight and reproducible framework for evaluating the pre-trained YOLO11s object detector in daylight and simulated night-time scenarios without any further retraining of the model. A typical road-scene frame from an actual daytime driving video was used to produce a corresponding simulated nighttime image with the same road geometry, vehicle locations, and camera viewpoint. The vehicle detection was preceded by enhancing the nighttime image using Contrast Limited Adaptive Histogram Equalization (CLAHE) and Gamma Correction. Detection behavior was evaluated by the number of detected vehicles, confidence-score statistics, confidence-level distribution, confidence ranking, scatter visualization, confidence-weighted heatmaps, and exploratory Pearson correlation analysis. The experimental results showed that YOLO11s detected 10 vehicles during daylight conditions and 3 vehicles under simulated nighttime conditions, which is about 70% less detections under reduced illumination. This is a decrease, but the most visually prominent vehicle still had a high confidence score in both conditions, while a number of weaker detections, including a dark-colored vehicle, were lost in the simulated nighttime scene. The proposed framework provides a simple and reproducible way to study the effect of illumination on the detection behavior of pretrained YOLO11s and highlights the need for further validation on larger datasets and real nighttime driving scenes.