Abstract
Segmentation and classification of surfaces and objects in automotive radar imagery are key techniques to identify the passable region for path planning in autonomous driving. The end-to-end segmentation on automotive radar imagery is proposed in this paper, where the input B-scope automotive radar map is processed to output the segmented radar map with labeled area classes. The algorithm discussed in this paper is the extension of our previous published work [1], where we proposed two-stage segmentation processed including (i) pre-segmentation using watershed transformation (WT), and (ii) the region classification based on the Multivariate Gaussian Distribution (MGD) classifier and the extracted distribution features. In the current paper, we use the B-scope radar map representation to simplify the coordinate transformation procedure as compared to PPI image representation. Secondly to improve classification of low-return regions two-tier segmentation process is introduced, where after the first classification of regions of high return, the more subtle classification is made between classes of low returns. Radar test dataset, collected in outdoor driving scenarios and labeled according to optical ground truth is used for assessing the Jaccard similarity coefficient (JSC) performance of segmentation results, which show higher accuracy of classification than in our previous algorithm [1].
Original language | English |
---|---|
Title of host publication | 2021 18th European Radar Conference (EuRAD) |
Publisher | IEEE |
Pages | 265-268 |
Number of pages | 4 |
ISBN (Electronic) | 9782874870651 |
ISBN (Print) | 9781665447232 (PoD) |
DOIs | |
Publication status | Published - 2 Jun 2022 |
Event | 18th European Radar Conference, EuRAD 2021 - London, United Kingdom Duration: 5 Apr 2022 → 7 Apr 2022 |
Publication series
Name | European Radar Conference (EURAD) |
---|
Conference
Conference | 18th European Radar Conference, EuRAD 2021 |
---|---|
Country/Territory | United Kingdom |
City | London |
Period | 5/04/22 → 7/04/22 |
Bibliographical note
Publisher Copyright:© 2022 European Microwave Association (EuMA).
Keywords
- Automotive sensor
- image segmentation
- radar imaging
- surface classification
ASJC Scopus subject areas
- Artificial Intelligence
- Computer Networks and Communications
- Computer Science Applications
- Computer Vision and Pattern Recognition
- Signal Processing
- Instrumentation