Urban Scene Parsing with Images and Scan Data

PhD Thesis Proposal Defence


Title: "Urban Scene Parsing with Images and Scan Data"

by

Mr. Honghui Zhang


ABSTRACT:

Urban scene parsing, segmenting interested objects and identifying their 
categories in urban scenes, is a fundamental issue in many applications, like 
3D city modeling and autonomous vehicles navigation. Different from the general 
scene parsing task, urban scene parsing is a typical representative of the 
constrained scene parsing task, which is an active research area. In this 
proposal, we investigate the methods for the urban scene parsing task with 
images and scan data, from small scale to large scale.

With both images and scan data, we propose a novel joint image and scan data 
scene parsing method which can be applied in large scale urban scenes. The 
proposed method can automatically obtain necessary training data from the input 
data, which is usually obtained through manually labeling in previous work. 
With the automatically obtained training data, we use an associative 
Hierarchical CRF to jointly optimize the segmentation of images and scan point 
cloud simultaneously. With only images, we propose a nonparametric scene 
parsing method which exploits the partial similarity between images, and a 
parametric scene parsing method, the supervised label transfer method. The 
partial similarity based nonparametric method involves no training process and 
reduces the inference problem in the scene parsing to a matching problem. By 
contrast, the supervised label transfer method transforms the inference problem 
in the scene parsing to a supervised matching problem. Last but not least, we 
propose an efficient Iterative Passive-Aggressive learning algorithm to learn 
the parameters involved in the random field models to formulate the scene 
parsing task, from some given training data. The parameters are iteratively 
updated by solving a structured output optimization problem, sharing similar 
updating form as the projected sub-gradient methods but without using any 
predefined step-size. The proposed methods are evaluated and compared with some 
state-of-the-art methods on several public datasets and the real Google Street 
View data, with encouraging performance achieved.


Date:                   Friday, 27 April 2012

Time:                   3:00pm - 5:00pm

Venue:                  Room 5510
                         lifts 25/26

Committee Members:      Prof. Long Quan (Supervisor)
                         Dr. Chiew-Lan Tai (Chairperson)
 			Dr. Pedro Sander
 			Prof. Chi-Keung Tang


**** ALL are Welcome ****