Publicação
LiDAR based 3D object tracking for autonomous driving
| Resumo: | Technology has become essential for society’s every-day-life and with the recent increase in artificial intelligence’s interest, this area has gained more and more relevance for both people (e.g., due to the increasing number of users of personal assistants, such as Siri, Alexa and Google Assistant) and service providers (e.g., Google search engine and social networks’ recommendation algorithms to keep users busy and active on their platforms - Facebook, Youtube, TikTok, etc.). Nevertheless, artificial intelligence has been applied to many other areas, such as targeted advertising to specific users, cybersecurity, medicine, and the automobile industry. Although artificial intelligence has not been the perfect solution in the aforementioned applications, it has been responsible for several significant improvements in the last decade. For example, in the automobile industry, there are more and more companies offering solutions for autonomous vehicles, being Tesla the most notorious. This evolution was driven by several factors, including need and interest in improving road safety, growing traffic problems that exist due to the increase of vehicles circulating, more reliable sensors, and recent advances in various areas of artificial intelligence, such as object detection, semantic segmentation, and object tracking. These three areas are interconnected. However, they have different purposes - the first two (detection and segmentation) more related to static frame analysis (e.g., image based analysis), while object tracking is usually applied in dynamic environments (e.g., sequence of frames, such as a video) where its input is processed in order to track objects over time, allowing an intelligent system to be “aware” of its environment. That said, this dissertation aims to study and explore the applicability and feasibility, as well as to develop and implement an object tracker in the context of autonomous driving. Furthermore, it is also intended to make a benchmark with state-of-the-art approaches and identify their main limitations. The input data will be focused on Light Detection and Ranging (LiDAR) based 3D point cloud, as there are several datasets available, in particular KITTI [1], which, in addition to being widely used in the state-of-the-art, has also achieved positive results, even in real-time execution situations. However, these solutions usually require a lot of computational resources and, which can be a hurdle for its application in real-life settings. |
|---|---|
| Autores principais: | Figueiredo, André Sousa |
| Assunto: | Object tracking Self-driving vehicles LiDAR Deep learning Tracking de objetos 3D Condução autonóma |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | Technology has become essential for society’s every-day-life and with the recent increase in artificial intelligence’s interest, this area has gained more and more relevance for both people (e.g., due to the increasing number of users of personal assistants, such as Siri, Alexa and Google Assistant) and service providers (e.g., Google search engine and social networks’ recommendation algorithms to keep users busy and active on their platforms - Facebook, Youtube, TikTok, etc.). Nevertheless, artificial intelligence has been applied to many other areas, such as targeted advertising to specific users, cybersecurity, medicine, and the automobile industry. Although artificial intelligence has not been the perfect solution in the aforementioned applications, it has been responsible for several significant improvements in the last decade. For example, in the automobile industry, there are more and more companies offering solutions for autonomous vehicles, being Tesla the most notorious. This evolution was driven by several factors, including need and interest in improving road safety, growing traffic problems that exist due to the increase of vehicles circulating, more reliable sensors, and recent advances in various areas of artificial intelligence, such as object detection, semantic segmentation, and object tracking. These three areas are interconnected. However, they have different purposes - the first two (detection and segmentation) more related to static frame analysis (e.g., image based analysis), while object tracking is usually applied in dynamic environments (e.g., sequence of frames, such as a video) where its input is processed in order to track objects over time, allowing an intelligent system to be “aware” of its environment. That said, this dissertation aims to study and explore the applicability and feasibility, as well as to develop and implement an object tracker in the context of autonomous driving. Furthermore, it is also intended to make a benchmark with state-of-the-art approaches and identify their main limitations. The input data will be focused on Light Detection and Ranging (LiDAR) based 3D point cloud, as there are several datasets available, in particular KITTI [1], which, in addition to being widely used in the state-of-the-art, has also achieved positive results, even in real-time execution situations. However, these solutions usually require a lot of computational resources and, which can be a hurdle for its application in real-life settings. |
|---|