A Case of Study on Traffic Cone Detection for Autonomous Racing on a Jetson Platform

Abstract

Autonomous driving is a growing research line since the future of transportation depends to a great extent on it. Driving is highly dependant on the environment sensing system. Over the last decade, several detection architectures based on neural networks and monocular cameras have been proposed to address this task. However, adapting these proposals to a vehicle with limited resources remains a challenging problem. In our study, we propose a lightweight neural network to perform cone detection from a racing car. We also compare its performance against other popular state-of-the-art proposals on a resource constrained system. From the obtained results, we can conclude that our network outperforms the state-of-the-art works for our use case and it is less resource demanding.