Deep Learning Based Low Light Enhancements for Advanced Driver-Assistance Systems at Night
Pamudu Ranasinghe, Dilshan Muthukuda, Primash Morapitiya, Maheshi Buddhinee Dissanayake, H. K. I. S. Lakmal · 2023
In the modern automobile industry, full/partial autonomous driving has become a common phenomenon. These systems heavily rely on sensors, both visual and non-visual, such as cameras and LIDAR, to interpret their surroundings and aid the decision-making process. Recently, visual-only sensing devices, mainly cameras, have emerged as the primary sensor for obtaining an immersive feeling in autonomous driving. However, the performance of such systems is heavily constrained by fluctuating light intensity and shadows. The primary goal of the presented research is to develop an effective low-light image enhancement system that improves the perception and inter-pretability of images taken in low-light driving environments. The developed image enhancement architecture demonstrates remarkable capabilities, surpassing existing models in terms of real-time suitability, PSNR (Peak Signal-to-Noise Ratio), and SSIM (Structural Similarity Index) values. Additionally, the model has been specifically designed to have low resource usage, making it an ideal solution for deployment on low-end devices. With significantly lower inference time, the model is exceptionally well-suited for real-time operation in autonomous driving applications. Furthermore, this architecture provides an efficient and effective solution not only for autonomous driving but also for a range of applications in video/image processing and computer vision, such as surveillance systems, medical imaging, astronomy, underwater imaging, and the mining industry. Finally, the relatively short training time makes it an attractive option for researchers and developers. Overall, this architecture represents a significant advancement in low-light image enhancement and autonomous vehicles. A demo of the system and the source code of the proposed LLE_UNET can be found on https://github.com/pamudu123ILow_Light_Image_Enhancement.