Registration using Gaussian mixture map for localization
Modern methods for vehicle localization rely on receiving reflectivities of the road markings with light detection and ranging (LIDAR). These reflectivities from the road can deteriorate because of inclement weather conditions or poor road texture which will decrease the accuracy of localization. Th...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
2019
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/78477 |
| Summary: | Modern methods for vehicle localization rely on receiving reflectivities of the road markings with light detection and ranging (LIDAR). These reflectivities from the road can deteriorate because of inclement weather conditions or poor road texture which will decrease the accuracy of localization. The improvement in LIDAR technology has brought attention to research in the point set registration for point cloud data (PCD) which needs to be efficient and accurate to be implemented for self-driving cars. In this study, a popular registration approach has been implemented which converts the a priori point cloud map into Gaussian mixture models (GMM), which is 2.5D map with height values. This GMM approach is compared to traditional Iterative Closest Point (ICP) approach in terms of point-to-point distance accuracy and computation time. The robustness of the GMM approach is tested and compared with ICP for different Gaussian noise levels. |
|---|