Robust License Plate Detection in Low-Light Conditions: Enhancing Accuracy for Real-World Applications

License plate detection in low-light conditions faces various challenges such as poor visibility, noise, motion blur, and plate design variations. To address the mentioned issues, we propose a deep learning-based framework integrating image enhancement and object detection techniques. Our approach utilizes MIRNet for enhancing low-light images by restoring details, enhancing contrast, and reducing noise. This preprocessing step significantly improves image quality, ensuring better visibility of license plates. To address motion blur, we incorporate Deblur-GANv2, which effectively reconstruct sharp details using generative adversarial networks and offers efficiency for various applications. Once images are enhanced and deblurred, YOLOv10, an object detection model, precisely localizes license plates. Its advanced feature extraction and multi-scale detection techniques ensure accurate identification, even in challenging conditions. Experimental evaluations demonstrate that our framework significantly improves detection accuracy in low-light and motion-blurred scenarios. Testing across diverse datasets, including night time and high-speed motion, shows superior results compared to traditional methods. In conclusion, our solution, combining MIRNet, DeblurGANv2, and YOLOv10, provides a robust and dependable approach for license plate recognition.