A Cloud-based Continual Learning System for Road Sign Classification in Autonomous Driving

Artificial intelligence and deep learning have demonstrated highly promising results in challenging problems, such as autonomous or assisted driving. One challenge in the integration of these solutions in real-life applications, is that they often operate in a resource-constrained edge environment. Another important challenge is the ability of the AI system to adapt and expand its abilities in a constantly changing environment. Constant changes could potentially cause significant deterioration of a model's effectiveness, a phenomenon called Catastrophic Forgetting. In this paper, we propose a Continual Learning framework for efficient and continuous update of a road sign classification system for assisted or autonomous driving. Our proposition considers the limitations of edge computing and utilizes a cloud infrastructure. Test results show that the our proposition is capable of expanding an edge models knowledge in a stable manner.