A survey on deep learning for data integration in autonomous driving systems

PhD Qualifying Examination


Title: "A survey on deep learning for data integration in autonomous 
driving systems"

by

Mr. Likang WANG


Abstract:

Autonomous cars perceive and decide according to the data collected from 
sensors. A single sensor generally suffers from limited applicable 
scenarios due to weather conditions and field of view. Thus, various 
multi-sensor perception systems have been proposed to exploit advances 
from different sensors. Most recent solutions were constructed on deep 
learning considering its success in broad domains. In this paper, we 
comprehensively survey state-of-the-art multi-source integration 
techniques in the perception module of self-driving vehicles. We analyze 
integration frameworks by asking three questions: "what, when, and how" to 
integrate. Our novel taxonomy divides integration strategies into seven 
classes by checking whether they consider inputs from multiple views, 
modalities, and frames. Unlike previous reviews, we sort out the pros and 
cons of different integration operations. Based on our comprehensive 
observation of existing strategies, we propose a portrait of the "ideal" 
integration framework, which may guide further research.


Date:			Wednesday, 7 December 2022

Time:                  	2:00pm - 4:00pm

Venue:                  Room 4472
                         Lifts 25/26

Committee Members:	Prof. Lei Chen (Supervisor)
  			Prof. Ke Yi (Chairperson)
 			Dr. Wei Wang
 			Prof. Bo Li


**** ALL are Welcome ****