Selected Problems of High-Resolution Automotive Imaging Radar

Xiao, Yang ORCID: 0000-0003-2804-9060 (2023). Selected Problems of High-Resolution Automotive Imaging Radar. University of Birmingham. Ph.D.

Preview

Xiao2023PhD.pdf
Text - Accepted Version
Available under License All rights reserved.
Download (12MB) | Preview

Abstract

This thesis aims at two selected problems in the development of high-resolution au- tomotive imaging radar: 1) The feasibility of using sub-THz for the next generation of automotive radar; 2) The development of the physics-based image segmentation approach on the automotive radar imagery.

The wide range of feasibility studies on the use of sub-THz frequencies for auto- motive radar have been undertaken in the Microwave Integrated Systems Laboratory (MISL) at the University of Birmingham, and the candidate is in charge of the included study on the theoretical modelling and experimental verification of the attenuation through the vehicle infrastructures which is the first part of this thesis. The importance of this work is related to the fact that automotive radar is placed within the car infras- tructure. Therefore, it would be a potential show-stopper in the development of this innovation if attenuation within the car bumper or badge is prohibitively high. Both theoretical modelling and experimental measurement are conducted by considering the impact factors on the propagation properties of the sub-THz signal such as the incident angle, frequency, characteristic parameters of materials, and the thicknesses of infrastructure layers. The transmissivity of multilayered structure has been modelled and good agreement with the results of measurements was demonstrated, so that the developed approach can be used in further studies on propagation through car infrastruc- ture. The published results on transmissivity and complex permittivity of automotive paints are valuable for researchers in either field of THz technology or automotive radar.

The image segmentation on automotive radar maps aims at identifying the passable and impassable areas for path planning in autonomous driving. Contrary to traditional radar, radar clutter is regarded as the physical meaningful information, which can deliver valuable feature information for surface characterization, and enable the full scene reconstruction of automotive radar maps. The proposed novel segmentation algorithm is a hybrid method composed of pre-segmentation based on image processing methods, and the region classification using the multivariate Gaussian distribution (MGD) classifier developed based on the statistical distribution feature parameters of radar returns of various areas. Moving target indication (MTI) is implemented for the first time based on frame-to-frame context association. The end-to-end segmentation framework is therefore achieved robustly with good segmentation performance, and automatically without human intervention.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):