Abstract:

Abstract: Autonomous vehicles (AVs) are poised to revolutionize transportation, but their performance in low-light conditions, such as nighttime or poorly lit environments, remains a significant challenge. Object detection in these conditions is crucial for safe navigation but often suffers due to reduced visibility and sensor limitations. Traditional deep learning models require large datasets of well-lit images, making them less effective in low-light scenarios. This project explores the use of Federated Learning (FL) for enhancing object detection capabilities of autonomous vehicles under low-light conditions. By enabling collaborative learning across multiple vehicles without the need to share sensitive data, FL allows for the development of robust, privacy-preserving models that adapt to the diverse and dynamic environments encountered by AVs. Incorporating sensor fusion (combining data from cameras, LiDAR, and radar), Federated Learning can aggregate knowledge from various AVs to improve object detection algorithms' performance across different lighting conditions. This approach leverages the distributed nature of AV fleets to continually refine models in real time, adapting to new data and environmental factors. The proposed system aims to enhance the safety and reliability of autonomous vehicles by improving their ability to detect objects in challenging lighting environments, thereby contributing to safer autonomous driving in urban and rural settings.

Description:

.