Data-driven techniques in signal restoration and detection problems

With the abundance of data and affordable computational power, data-driven approaches have exploded in the last few years, solving various problems in the field of computer vision with performance exceeding human performance. In this work, we study how learned priors can be applied to the task of image restoration, without having to explicitly train for such a task. We also study how traditional signal processing chains for radars can be augmented with modern data-driven techniques. Our experiments showed that there is sufficient information in a radar heatmap to reliably identify the class of the reflecting object. Using an automatically labeled dataset, we were able to achieve a classification accuracy of over 85% in the indoor scenario and over 98% in the outdoor scenario. Our evaluation, performed on real world data, suggests that the complementary nature of radar and camera signals can be leveraged to reduce the lateral error by 15% when applied to object detection Finally, we also propose a novel method for combining multiple sensor observations in a learned feature space, demonstrating robustness to sensor failure.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2022

Author dc:creator

• Lim, Teck-Yian

Contributors dc:contributor

• Do, Minh N
• Schwing, Alexander G
• Forsyth, David A
• Gupta, Saurabh
• Wang, Yuxiong

• FMCW radar
• image restoration
• sensor fusion

Statement dc:rights

• Copyright 2022 Teck-Yian Lim

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/115945

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F115945