Vision-based lateral position improvement of RADAR detections
Authors:
- 1 September 2012
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- No. 21530009,p. 90-97
Abstract
Detecting vehicles ahead by sensors mounted on an ego-vehicle is an essential element of Advanced Driver Assistance Systems (ADAS). Although millimeter-wave radar sensors are very robust at detecting vehicles, the lateral position resolution is very low. However, a more precise lateral position would not only improve the performance of existing ADAS, but it would also enable a wide range of additional applications. In this paper, we propose a method for improving the lateral position of radar sensor detections by fusing them with a vision sensor. Our method is based on a robust symmetry axis detection of vehicles' rear sides. In contrast to the typical approach of detecting symmetry by means of edges, we use a patch-matching-based approach together with a robust cost function. The initial lateral position of the radar detection is improved by conducting a local symmetry search in the image region corresponding to the objects detected by the radar. The experimental results show that the proposed method is able to improve the lateral position accuracy by a factor of seven. Furthermore, additional consistency checks discover miss-detections of the symmetry axis, thus ensuring that the lateral position never drops below the accuracy of the radar detection.Keywords
This publication has 8 references indexed in Scilit:
- A Two-Stage Correlation Method for Stereoscopic Depth Estimation2010
- Multi-resolution vehicle detection using artificial vision2004
- Real-time vehicle following through a novel symmetry-based approach2002
- Stereo vision-based vehicle detection2002
- Visual perception of obstacles and vehicles for platooningIEEE Transactions on Intelligent Transportation Systems, 2000
- Symmetry-based recognition of vehicle rearsPattern Recognition Letters, 1991
- Box-filtering techniquesComputer Graphics and Image Processing, 1981
- A New Approach to Linear Filtering and Prediction ProblemsJournal of Basic Engineering, 1960