Object tracking and recognition

MPhil Thesis Defence


Title: "Object tracking and recognition"

By

Miss Shengnan CAI


Abstract

This paper concludes our previous works on visual tracking and place 
recognition. The categorization is yielded based on features and algorithms the 
diverse object recognition methods utilize.

In place recognition, we study the problem of recognizing man-made objects and 
present a novel affine-invariant feature, Low-rank SIFT, which exploits the 
regular appearance property in man-made objects. After an analysis on various 
features representing the scene, we propose a novel feature which achieves full 
affine invariance without needing to simulate over affine parameter space. We 
rectify local patches by converting them to their low-rank forms to achieve 
skew invariance, and perform the way similar to conventional SIFT to resolve 
rotation, translation and scaling ambiguity. The main contributions lie in 
two-fold: our method seeks to leverage low-rank prior to estimate affine 
parameters for local patches directly and we propose a fast algorithm to 
compute such parameters by introducing the Low-rank Integral Map.

In visual tracking, we propose two approaches. One utilizes generalized 
part-based appearance model and structure-constrained motion model as 
auxiliary. The appearance of the target object is modeled by the proposed 
generalized part-based appearance model, which combines the appearance of 
different parts of the target object, adaptively updated by an efficient 
structure learning scheme based on the online Passive-Aggressive algorithm. By 
integrating the confidence scores of multiple parts, mutual compensation is 
realized, significantly enhances the robustness of our method against the 
structure deformation and partial occlusion during the tracking. In addition, 
we enhance the performance of our tracker by using a motion model. It employs a 
structure-constrained rule, that is, the change on the structure of the target 
object between consecutive frames is small. Another tracking method leverages 
layered detection that combines detection on two independent layers in a 
unified tracking-by-detection framework, one layer on the global level and the 
other on patch. Based on the bounding box representation for the object of 
interest, the detection on the global level is formulated with the structured 
prediction framework that is superior for distinguishing the background and 
object of interest during the tracking. For the patch level detection, an 
efficient patch level detector which is robust against the sampling error 
during the online updating is proposed. With the patch level detector, 
confidence estimation for the background and object of interest on patch level 
is carried out for the tracking. Comprehensive evaluations of our methods are 
conducted on a public benchmark for object tracking, and the experiment result 
shows that the proposed method using layer detection for object tracking 
outperforms state-of-the-art algorithms, with observable improvement 
demonstrated.


Date:			Tuesday, 17 March 2015

Time:			10:30am - 12:30pm

Venue:			Room 3501
 			Lifts 25/26

Committee Members:	Prof. Long Quan (Supervisor)
 			Prof. Chiew-Lan Tai (Chairperson)
 			Dr. Huamin Qu


**** ALL are Welcome ****