Home // VEHICULAR 2017, The Sixth International Conference on Advances in Vehicular Systems, Technologies and Applications // View article

Application of Deep Learning to Route Odometry Estimation from LiDAR Data

Authors:
Miguel Clavijo
Francisco Serradilla
Jose-E Naranjo
Felipe Jiménez
Alberto Díaz

Keywords: deep learning; autonomous vehicle; odometry; LiDAR

Abstract:
The Deep Learning techniques are a powerful tool to support the development of all sorts of information classification or processing techniques within the area of intelligent vehicles, since they are able to emulate the performance of the human brain when learning from experience. Specifically, the technique of Convolutional Neural Networks (CNN) has been successfully used in applications for classification and localization of pedestrians and obstacles on the road. However, CNN allow not only classification and pattern learning, but can be used for regression or modeling, like other kind of classical neural networks. The fundamental difference of both applications is that, while in classification the values of the network output are usually discrete, in regression or modeling applications the network can generate a continuous output with real numbers, allowing it to emulate the output of any type of system that is presented in the training set, with all its associated advantages, such as generalization and correct characterization of situations that have not learned explicitly. This paper presents an application of CNN for modeling in Intelligent Vehicles field, whose objective is to calculate the navigation parameters of a vehicle from the information supplied by a 3D LiDAR mounted on a vehicle that circulates in urban areas. Specifically, the developed CNN is able to calculate the speed and heading of a vehicle circulating in real time from the distance data supplied by the LiDAR sensor. The results show that the network is able to learn to calculate the speed and the yaw rate from the identification of the characteristic points of the environment, providing data that can be used to support the navigation of the vehicles.

Pages: 60 to 65

Copyright: Copyright (c) IARIA, 2017

Publication date: July 23, 2017

Published in: conference

ISSN: 2327-2058

ISBN: 978-1-61208-573-9

Location: Nice, France

Dates: from July 23, 2017 to July 27, 2017