Publicação
Roadside Boundary Perception for Autonomous Driving using Machine Learning techniques on Radar Reflections
| Resumo: | Autonomous Driving is considered a vital part in the development of the future of urban transportation systems. Driving assistance technology through built-in cameras and radar sensors are realized for assisted parking systems or to prevent potential crash accidents. Recent research and development on highly automated driving is aimed to ensure stability, improve robustness and enhance safety of driving assistance systems in lane recognition or lane departure avoidance through radar perception. For optimization and refinement of radar perception, an environment detection of stationary or non-stationary road objects is realized. Techniques such as Machine Learning and Deep Learning are used to perform classification on road objects to predict their respective localization. Radar sensors installed in cars are developed with a high spatial resolution and computing power. However, the rising amount of non-usable data stands in correlation with the increasing resolution. Despite the progress made in the development of advanced driver assistance system (ADAS), an accurate perception of the vehicle’s surroundings in 3D point clouds is considered to be a challenge in the research of autonomous vehicle software systems. The objective of this thesis is to implement an artificial neural network for the intelligent filtering of the perceived radar reflections, represented as point clouds. The neural network should improve the perception of roadside boundaries, such as identifying guardrails or roadedge objects, by extracting relevant data and enabling efficient further processing. This project is developed within the scope of an internship at Bosch Group Germany. Given the implementation of an improved filtering of roadside boundary objects through a neural network and the reduction of the misclassification rate of guardrail locations, the perception of the autonomous vehicle software system is improved and the focus is centered on processing relevant data more efficiently. |
|---|---|
| Autores principais: | Leuthner, Maren Corina |
| Assunto: | Autonomous Driving Perception Machine Learning Classification Neural Networks Condução autónoma Percepção Aprendizagem mecânica Classificação Redes Neurais |
| Ano: | 2021 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | Autonomous Driving is considered a vital part in the development of the future of urban transportation systems. Driving assistance technology through built-in cameras and radar sensors are realized for assisted parking systems or to prevent potential crash accidents. Recent research and development on highly automated driving is aimed to ensure stability, improve robustness and enhance safety of driving assistance systems in lane recognition or lane departure avoidance through radar perception. For optimization and refinement of radar perception, an environment detection of stationary or non-stationary road objects is realized. Techniques such as Machine Learning and Deep Learning are used to perform classification on road objects to predict their respective localization. Radar sensors installed in cars are developed with a high spatial resolution and computing power. However, the rising amount of non-usable data stands in correlation with the increasing resolution. Despite the progress made in the development of advanced driver assistance system (ADAS), an accurate perception of the vehicle’s surroundings in 3D point clouds is considered to be a challenge in the research of autonomous vehicle software systems. The objective of this thesis is to implement an artificial neural network for the intelligent filtering of the perceived radar reflections, represented as point clouds. The neural network should improve the perception of roadside boundaries, such as identifying guardrails or roadedge objects, by extracting relevant data and enabling efficient further processing. This project is developed within the scope of an internship at Bosch Group Germany. Given the implementation of an improved filtering of roadside boundary objects through a neural network and the reduction of the misclassification rate of guardrail locations, the perception of the autonomous vehicle software system is improved and the focus is centered on processing relevant data more efficiently. |
|---|