Work

"LiDAR deep learning lane making detection."

Year: 2024
Type: dissertation
Abstract: Autonomous driving has become increasingly important for drivers and car manufacturers in recent years. This demand has forced development not only in new electronic and hardware components, but also in the software, and it has a huge impact on autonomous driving. Without the use of artificial intelligence models, autonomous driving would not be possible at all, and with the increasing creation of new equipment, it is necessary to develop a new deep learning models for this same equipment. In this thesis, we will solve the problem of how to build a Framework to support the development of new deep learning models for detecting road lines using data obtained from LiDAR in the form of a Point Cloud. The modules that have been developed will be presented, such as the data extraction module and processing module, the training module, and the prediction module. We will present all the methodologies and technical decisions that served as the basis and the decision support for the implementation of a model based on PointPillars Net, the construction of a new Head Net and how it fits into the model. The methods used to quantify training results and model prediction after training will be demonstrated, as well as the methods used to calculate each of the components that support the Framework or PointPillars model. We will demonstrate the Framework's final structure and how it was used to help extract the data from the datasets, train and predict the model. We will demonstrate the current and future applications that the Framework could have in the development of new lane detection models.
Source: CiencIPCA
Author: Araújo, Domingos Manuel Costa
Language: English
Open Access status: green
OpenAlex ID: W6992438999
Cites: 82
Cited by:
Related to: 10