| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 4.82 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A aplicação de Inteligência Artificial (IA) no setor automóvel tem vindo a ser uma mais valia principalmente quando se fala em segurança rodoviária. A ligação entre o Homem e Máquina através da IA está vindo a ser mais fortalecida desde os anos 50 do século XX e a partir do século XXI, tem-se visto um crescimento tecnológico exponencial, estando disponíveis atualmente uma infinidade de aplicações em Machine Learning (ML) e Deep Learning (DL) nos vários setores existentes, desde o setor industrial ao setor da saúde. No sub-setor que é a condução autónoma, a aplicação de algoritmos é feita principalmente para a segurança rodoviária, isto é, mapeamento do ambiente, deteção de objetos, pessoas, entre outros, leitura de sinalizações e previsão de trajetórias. Sendo também utilizada na categoria de conetividade como iteração com o condutor na leitura de gestos ou conetividade entre veículos, conhecida como Vehicle-to-everything (V2X). Já com uma certa maturidade deste conceito no setor automóvel, são várias as marcas e grupos que pretendem aplicá-lo em tarefas mais complexas, levando um dos propósitos da IA, a condução autónoma, para um próximo patamar. Esta dissertação pretende assim dar entrada à condução autónoma com o processamento de diferentes dados veiculares e implementação de técnicas de DL para a previsão de trajetórias em MATLAB. Com este objetivo, foi primeiramente feita uma revisão bibliográfica sobre o tema, depois uma análise dos dados Next Generation Simulation (NGSIM) fornecidos pelo Departamento de Transporte dos Estados Unidos e desenvolvimento dos algoritmos e, por fim, realização de testes e comparação do desempenho dos vários algoritmos utilizados. Desta forma, e sendo a previsão de trajetórias, em MATLAB, um tópico ainda com pouca aplicação de inteligência artificial, comparado com a classificação de imagens, pretende-se fornecer uma base de projeto para a entrada nesta área da condução autónoma.
The application of artificial intelligence in the automotive sector has been an asset, especially when it comes to road safety. The link between Man and Machine through artificial intelligence has been strengthened since the 50s of the 20th century and from the 21th century, technological advances have been seen with a exponentially growth, with currently a countless applications in ML and DL in the different existing sectors, from the industrial sector to the health sector. In the autonomous driving sub-sector, the application of algorithms is mainly done for road safety, that is, mapping the environment, detecting objects, people, among others, reading signs and predicting trajectories, being also used in category of connectivity such as iteration with the driver in reading gestures or connectivity between vehicles, known as V2X. With a certain maturity of this concept in the automotive sector, there are several brands and groups that intend to apply it in more complex tasks, taking one of the purposes of IA, autonomous driving, to the next level. This dissertation therefore intends to introduce autonomous driving with the processing of different vehicular data and implementation of DL techniques for the prediction of trajectories in MATLAB. With this goal, a bibliographic review on the subject was first carried out, then an analysis of the NGSIM data provided by the US Department of Transportation and development of the algorithms and, finally, tests and comparison of the performance of the different algorithms used. In this way, and since the trajectory prediction, in MATLAB, is still a topic with low number of applications of artificial intelligence, compared to image classification, it is intended to provide a project basis for entry into this area of autonomous driving.
The application of artificial intelligence in the automotive sector has been an asset, especially when it comes to road safety. The link between Man and Machine through artificial intelligence has been strengthened since the 50s of the 20th century and from the 21th century, technological advances have been seen with a exponentially growth, with currently a countless applications in ML and DL in the different existing sectors, from the industrial sector to the health sector. In the autonomous driving sub-sector, the application of algorithms is mainly done for road safety, that is, mapping the environment, detecting objects, people, among others, reading signs and predicting trajectories, being also used in category of connectivity such as iteration with the driver in reading gestures or connectivity between vehicles, known as V2X. With a certain maturity of this concept in the automotive sector, there are several brands and groups that intend to apply it in more complex tasks, taking one of the purposes of IA, autonomous driving, to the next level. This dissertation therefore intends to introduce autonomous driving with the processing of different vehicular data and implementation of DL techniques for the prediction of trajectories in MATLAB. With this goal, a bibliographic review on the subject was first carried out, then an analysis of the NGSIM data provided by the US Department of Transportation and development of the algorithms and, finally, tests and comparison of the performance of the different algorithms used. In this way, and since the trajectory prediction, in MATLAB, is still a topic with low number of applications of artificial intelligence, compared to image classification, it is intended to provide a project basis for entry into this area of autonomous driving.
Description
Keywords
Previsão de Trajetórias NGSIM Inteligência Artificial Machine Learning Deep Learning MATLAB Trajectory Prediction Artificial Intelligence Machine Learning