Vehicle detection using radar and vision with noisy data
<p>Autonomous vehicles operating in on-road environments need to be aware of the presence and motion of other actors in the environment. The detection of other vehicles is therefore a key component of an autonomous driving system. This thesis investigates how the highly complementary character...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
2019
|
| _version_ | 1797085328478044160 |
|---|---|
| author | Chadwick, SPA |
| author2 | Newman, P |
| author_facet | Newman, P Chadwick, SPA |
| author_sort | Chadwick, SPA |
| collection | OXFORD |
| description | <p>Autonomous vehicles operating in on-road environments need to be aware of the presence and motion of other actors in the environment. The detection of other vehicles is therefore a key component of an autonomous driving system. This thesis investigates how the highly complementary characteristics of radar and vision can be combined to tackle the problem. It first discusses the issues surrounding the implementation of a multi-modal sensing system, including calibration and synchronisation. It then shows how the range and direct velocity measurements of radar can be used to improve the performance of current image-based detectors when detecting small, distant vehicles. Given the importance of labelled data to the training of modern object detectors, it introduces a method for automatically labelling a dataset utilising cameras of different focal lengths. The impact of errors in the training labels on the performance of the trained detector is then demonstrated. To mitigate that impact, a technique for training object detectors using noisy labels is introduced. Finally, the use of the radar as a labelling source is considered along with a novel process for training a detector using the radar labels.</p> |
| first_indexed | 2024-03-07T02:07:24Z |
| format | Thesis |
| id | oxford-uuid:9f72a271-8c3c-4f6b-8ee3-5c7c6284d892 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T02:07:24Z |
| publishDate | 2019 |
| record_format | dspace |
| spelling | oxford-uuid:9f72a271-8c3c-4f6b-8ee3-5c7c6284d8922022-03-27T00:57:56ZVehicle detection using radar and vision with noisy dataThesishttp://purl.org/coar/resource_type/c_db06uuid:9f72a271-8c3c-4f6b-8ee3-5c7c6284d892EnglishHyrax Deposit2019Chadwick, SPANewman, P<p>Autonomous vehicles operating in on-road environments need to be aware of the presence and motion of other actors in the environment. The detection of other vehicles is therefore a key component of an autonomous driving system. This thesis investigates how the highly complementary characteristics of radar and vision can be combined to tackle the problem. It first discusses the issues surrounding the implementation of a multi-modal sensing system, including calibration and synchronisation. It then shows how the range and direct velocity measurements of radar can be used to improve the performance of current image-based detectors when detecting small, distant vehicles. Given the importance of labelled data to the training of modern object detectors, it introduces a method for automatically labelling a dataset utilising cameras of different focal lengths. The impact of errors in the training labels on the performance of the trained detector is then demonstrated. To mitigate that impact, a technique for training object detectors using noisy labels is introduced. Finally, the use of the radar as a labelling source is considered along with a novel process for training a detector using the radar labels.</p> |
| spellingShingle | Chadwick, SPA Vehicle detection using radar and vision with noisy data |
| title | Vehicle detection using radar and vision with noisy data |
| title_full | Vehicle detection using radar and vision with noisy data |
| title_fullStr | Vehicle detection using radar and vision with noisy data |
| title_full_unstemmed | Vehicle detection using radar and vision with noisy data |
| title_short | Vehicle detection using radar and vision with noisy data |
| title_sort | vehicle detection using radar and vision with noisy data |
| work_keys_str_mv | AT chadwickspa vehicledetectionusingradarandvisionwithnoisydata |