Automated tagging of safety cases for autonomous vehicles in Asian megacities
With the emerging technologies and fast-moving world, the cars nowadays are getting increasingly automated and thus use powerful algorithms. The scenarios, an automated vehicle has to deal with are different when comparing Asian countries with, e.g. European countries. For scenario analysis in Singa...
Saved in:
| 主要作者: | |
|---|---|
| 其他作者: | |
| 格式: | Theses and Dissertations |
| 語言: | English |
| 出版: |
2018
|
| 主題: | |
| 在線閱讀: | http://hdl.handle.net/10356/73124 |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 機構: | Nanyang Technological University |
| 語言: | English |
| 總結: | With the emerging technologies and fast-moving world, the cars nowadays are getting increasingly automated and thus use powerful algorithms. The scenarios, an automated vehicle has to deal with are different when comparing Asian countries with, e.g. European countries. For scenario analysis in Singapore, research cars collect vehicle data comprising information about the car as well as image data. The extraction of useful information out of this data is a challenge. In this thesis, the research is mainly related to automatic labelling of data example by applying cluster analysis, Computer vision algorithm etc. The goal of this work is to expand, examine and consider algorithms which allow to study the images and allocate them to predefined labels. This algorithm suggested in this thesis is especially suitable for speed hump detection, static traffic regulation objects which force drivers to reduce the speed. More than 16TB of data was collected to be analysed them. An algorithm for detecting speed hump via machine learning has been discussed in [1]. This thesis formulates an innovative approach by tagging the data to detect speed hump in the images collected from the front camera located in research cars by using Computer vision algorithm. Detection of speed hump in the image is not a simple task. Various feature extraction technique such as HoG / Haar / LBP need to be implemented to extract the features. Among all these feature extraction techniques, HoG was found to be the best. This has been followed by applying Cascade object detection technique which works on sliding window approach to detect the ROI (Region of Interest), which in our case are speed hump. In this thesis, the Computer vision model is trained and optimized for accuracy by tuning various parameters which are present in the algorithm. The average precision of speed hump detection via Computer vision algorithm was found to be approximately 92%. |
|---|