Share

Export Citation

APA
MLA
Chicago
Harvard
Vancouver
BIBTEX
RIS
Universitas Hasanuddin
Research output:Contribution to journalArticlepeer-review

Improving Object Detection Accuracy in Low-Light Conditions Through Integration of Contrast Limited Adaptive Histogram Equalization into Faster RCNN Architecture

Pulung D.

Proceedings of the 23rd IEEE International Conference on Computer Applications Icca 2026

Published: 2026

Abstract

Low-light conditions significantly degrade object detection performance in urban traffic monitoring due to insufficient contrast and poor feature visibility. To address this issue, this study integrates Contrast Limited Adaptive Histogram Equalization (CLAHE) into the Faster R-CNN architecture as a preprocessing stage. CLAHE is applied in the LAB color space with a clip limit of 5.0 and a tile grid size of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$8 \times 8$</tex> to enhance luminance while suppressing noise amplification. Experiments were conducted on a custom dataset of 661 nighttime traffic images collected in Rantepao City using an iPhone <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$14(1080 \mathrm{p}, 30 \text{fps})$</tex>. The proposed CLAHE-enhanced Faster RCNN achieves an mAP of 0.628, representing a 102.6 % improvement over the baseline model (mAP 0.310). The Average Precision for cars increases to 0.914 (46.2 %), while motorcycles reach <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$0.709(164.6 \%)$</tex>. Training convergence is accelerated by 2.7, achieving loss 0.3 at epoch 10, compared to stagnation above 0.8 in the baseline. Although computational time increases by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$8.5 \%(9,220$</tex> second vs. 8,497 second), histogram-based analysis confirms a 157 % increase in mean intensity, 87 % improvement in contrast, and 20 % increase in entropy, indicating enhanced feature discriminability. These results demonstrate that integrating CLAHE into Faster RCNN provides a robust and effective solution for vehicle detection in extreme low-light conditions, with strong potential for night-time surveillance and autonomous driving applications.

Other files and links

Fingerprint

Computer scienceSciences
Artificial intelligenceSciences
Object detectionSciences
Contrast (vision)Sciences
Computer visionSciences
Pattern recognition (psychology)Sciences
Adaptive histogram equalizationSciences
Object (grammar)Sciences
HistogramSciences
Equalization (audio)Sciences
Viola–Jones object detection frameworkSciences
Noise (video)Sciences
Histogram equalizationSciences
Feature (linguistics)Sciences
ArchitectureSciences
Channel (broadcasting)Sciences
Cognitive neuroscience of visual object recognitionSciences