Detecção de Faixa Baseado em Aprendizado Profundo para Seguimento de Caminho Visual

Abstract

Detecting a path or trajectory that a vehicle should follow is one of the most important tasks in autonomous robot navigation. Over the past few years, some learning-based works has stood out more than traditional computer vision techniques in detecting paths. This paper presents a transfer-learning-based approach to solve the lane line detection problem in the context of visual path following by using a residual factorized convolutional neural network trained using an approximation of IoU as the loss function. Experimental results show a promising model that can detect paths even in the presence of discontinuities and light variations. The proposed model architecture also strikes a good balance between accuracy and efficiency, making the system suitable for real-time applications.