Learning based terrain traversability analysis
Autonomous mobile robot has been becoming a promising way for some human-risky tasks, suck like search and rescue, mining, planetary exploration, medical service and warehouse logistics. To assure the safety of the system, the perception ability of terrain traversability is playing a significant rol...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis-Master by Coursework |
| Language: | English |
| Published: |
Nanyang Technological University
2020
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/141062 |
| _version_ | 1811694543455649792 |
|---|---|
| author | Fang, Hao Yu |
| author2 | Wang Dan Wei |
| author_facet | Wang Dan Wei Fang, Hao Yu |
| author_sort | Fang, Hao Yu |
| collection | NTU |
| description | Autonomous mobile robot has been becoming a promising way for some human-risky tasks, suck like search and rescue, mining, planetary exploration, medical service and warehouse logistics. To assure the safety of the system, the perception ability of terrain traversability is playing a significant role by taking into the slope, obstacle and terrain classes into consideration. The goal of this research is aiming to develop a learning-based method to efficiently and precisely percept the terrain for a mobile robot in the outdoor environment.
This project aims to obtain outdoor environment information directly through point cloud segmentation by deep learning method. It avoids the impact of harsh outdoor lighting conditions (too much light or too low light) on the images collected by the camera sensor. However, it is tedious to obtain the label of the point cloud dataset according to the manual labeling method. Therefore, an efficient method is also needed to obtain the label of the point cloud.
This task uses the information based on image semantic segmentation as the reference standard for point cloud semantic information. The full convolutional network trained on the public dataset is used to obtain image semantic segmentation information. Then, the image semantic information is fine-tuned to obtain a segmented image with a semantic information accuracy of 95% or more. the point cloud semantic segmentation labels are obtained through the correspondence between the point cloud and image mapping. Finally, testing accuracy after training on training dataset is 88.1%. |
| first_indexed | 2024-10-01T07:09:15Z |
| format | Thesis-Master by Coursework |
| id | ntu-10356/141062 |
| institution | Nanyang Technological University |
| language | English |
| last_indexed | 2024-10-01T07:09:15Z |
| publishDate | 2020 |
| publisher | Nanyang Technological University |
| record_format | dspace |
| spelling | ntu-10356/1410622023-07-04T16:32:51Z Learning based terrain traversability analysis Fang, Hao Yu Wang Dan Wei School of Electrical and Electronic Engineering EDWWANG@ntu.edu.sg Engineering::Electrical and electronic engineering Autonomous mobile robot has been becoming a promising way for some human-risky tasks, suck like search and rescue, mining, planetary exploration, medical service and warehouse logistics. To assure the safety of the system, the perception ability of terrain traversability is playing a significant role by taking into the slope, obstacle and terrain classes into consideration. The goal of this research is aiming to develop a learning-based method to efficiently and precisely percept the terrain for a mobile robot in the outdoor environment. This project aims to obtain outdoor environment information directly through point cloud segmentation by deep learning method. It avoids the impact of harsh outdoor lighting conditions (too much light or too low light) on the images collected by the camera sensor. However, it is tedious to obtain the label of the point cloud dataset according to the manual labeling method. Therefore, an efficient method is also needed to obtain the label of the point cloud. This task uses the information based on image semantic segmentation as the reference standard for point cloud semantic information. The full convolutional network trained on the public dataset is used to obtain image semantic segmentation information. Then, the image semantic information is fine-tuned to obtain a segmented image with a semantic information accuracy of 95% or more. the point cloud semantic segmentation labels are obtained through the correspondence between the point cloud and image mapping. Finally, testing accuracy after training on training dataset is 88.1%. Master of Science (Computer Control and Automation) 2020-06-03T11:24:36Z 2020-06-03T11:24:36Z 2020 Thesis-Master by Coursework https://hdl.handle.net/10356/141062 en application/pdf Nanyang Technological University |
| spellingShingle | Engineering::Electrical and electronic engineering Fang, Hao Yu Learning based terrain traversability analysis |
| title | Learning based terrain traversability analysis |
| title_full | Learning based terrain traversability analysis |
| title_fullStr | Learning based terrain traversability analysis |
| title_full_unstemmed | Learning based terrain traversability analysis |
| title_short | Learning based terrain traversability analysis |
| title_sort | learning based terrain traversability analysis |
| topic | Engineering::Electrical and electronic engineering |
| url | https://hdl.handle.net/10356/141062 |
| work_keys_str_mv | AT fanghaoyu learningbasedterraintraversabilityanalysis |