Tuan Hue, THI
Graduate Student , School of Computer Science and Engineering
University of New South Wales, Sydney, Australia
Level 4, 223 Anzac Parade, Kensington, NSW, 2033, Australia
Phone: +61 2 8306 0447
Mobile : 0404 659 624
Email: huetuan1984@gmail.com
Web: http://huetuan.net
Project Title Implicit Motion-Shape Model and Approaches to Visual-based Video Searching
Joint work Tuan Hue THI, Jian Zhang (NICTA-Sydney), Li Cheng (TTI-Chicago), Li Wang (SEU-China), and Shin'ichi Satoh (NII-Tokyo)
Project Description
In this project we aim to develop a complete visual-based video search using its motion content. Traditional text-based video search engines rely solely on annotated text description of a video to carry out similarity search. Due to the fast growth of video domain, manual annotation are normally labor expensive and can also be errorprone. Our motivation is to develop a system that uses motion content inside each video to carry out matching task, then rank video similarity according to the returned matching values. We are particularly interested in two aspects of the matching algorithm, firstly, it should work for realistic cases with cluttered backgrounds, secondly, it should be generic enough to work on any kind of action. Taking these into account, we turn our focus to Invariant Local Features and Nonparametric Implicit Shape Model.
Snapshots
Project Title Semi-supervised Human Action Recoginition and Localization using Spatially and Temporally Integrated Local Features
Joint work Tuan Hue THI, Jian Zhang (NICTA-Sydney), Li Cheng (TTI-Chicago), Li Wang (SEU-China), and Shin'ichi Satoh (NII-Tokyo)
Project Description
This projects develops a framework in recognizing and localizing human action in video sequences using weakly supervised approach. Local space-time features are detected from video shots and represented in histogram vector of oriented gradients and flows. A Sparse Bayesian Kernel classification model is built to represent the compact characteristics of supervised data and adaptive to unknown data, which purpose is to label each local feature according to relevant class of action. Group Constraints among local features and Markov Chain Monte Carlo sampling are augmented into the model via data association to boost up the performance in accuracy and processing time. The labeling assignment results are first passed into a non-linear Support Vector Machine to decide the class action of the whole video shot. Then the same label values are fed into a Conditional Random Field to propagate label information among neighboring regions, hence, accurately locate the event areas. Testing of this proposed weakly trained model on the classical KTH dataset, the realistic Hollywood Human Action dataset and the challenging TRECVID event detection dataset has yielded the comparable results to most of state-of-the-art fully supervised techniques.
Snapshots
Project Title Night-time Vehicle Detection and Tracking in Traffic Surveillance (NICTA)
Supervisors Prof Hung Nguyen (UTS), Mr. Sijun Lu, Dr. Getian Ye and Dr. Jian Zhang (NICTA)
Project Description
The aim of this project is to develop a system designed to gather useful statistics about road conditions at night. The data is collected from cameras and processed to automatically detect the lane lines, calibrate the camera data, detect vehicles using machine learning methods, track vehicles over multiple frames, and gather and display statistics such as lane coverage and vehicle speed.
Snapshots

 
Project Title Urban Search and Rescue Robot (Robocup) (CAS)
Supervisors Professor Gamini Dissanayake, Dr Jonathan Paxman, and Dr Jaime Valls Miro
Project Description
In an urban search and rescue scenario, detecting the locations of survivors and then recovering them from a collapsed building is one of the biggest challenges faced by emergency response personnel. The environment can be unstable and difficult to negotiate while survivors trapped need to be rescued within a short time frame. Use of a robot or a team of robots to assist human rescuers in such
situations is one of the areas where robotics research can be of great benefit to humanity. While significant progress has been made and a variety of robots have been used during many recent disasters, much research is needed to achieve
the objective of deploying a team of autonomous robots for urban search and rescue.The team participated in an international competition on urban search
and rescue (RoboCup Rescue) held in Bremen, Germany in June 2006 where HOMER was placed second in the autonomy challenge.
Snapshots

Project Title 3D Scene Reconstruction using Stereo camera Captures (author of S3dViewer and Composer) (CAS)
Supervisors Dr Jaime Valls Miro
Project Description
Reconstruction of 3D Scene based on images taken from stereo camera using S3d datatype as standard data structure. The outcome of this project was developed to build two softwares known as S3DViewer and S3DComposer
Snapshots

Project Title Managing Knowledge in Web-Driven Organisational Evolution (UTS)
Supervisors Prof David Lowe and Prof Didar Zowghi
Project Description

This project aims understand the interplay between business models and Web systems and provide support for managing it effectively. We will develop models that link system designs to business impacts, and provide techniques and research tools to assess these impacts. Possible approach is to link to the agile domain modelling project

Snapshots

 
Project Title Vision based SLAM using Support Vector Machine (CAS)
Supervisors Professor Gamini Dissanayake and Dr Jaime Valls Miro
Project Description
This project is an attempt to depart from traditional sensors such as laser rangefinders in order to gain the many benefits of nature-inspired information-rich 3D vision sensors. Whilst this makes the system fully observable in that the sensor provide enough information (range and bearing) to compute the full 2D estate of the observed landmarks from a single position, it is also true that depth information is difficult to rely on, particularly on measurements beyond a few meters (in fact the full 3D estate is observable, but here robot motion is constrained to 2D and only the 2D problem is considered). The work being carried out is focusing on two aspects:
- a partially measurable SLAM perspective in that only landmark bearing from one of the cameras is employed in the fusion estimation. Range information estimates from the stereo pair is only used during map building in the landmark initialization phase in order to provide a reasonably accurate initial estimate. An additional benefit of this approach lies in the data association aspect of SLAM. The availability of powerful feature extraction algorithms from the vision community, such as SIFT, permits a more flexible SLAM implementation separated from feature representation, extraction and matching, essentially carrying out matching with minimal recourse to geometry.
- a fully observable SLAM where range estimates are also incorporated in the filter
Snapshots

Homepage Resume Latex CV Projects Photo Albums Research Ideas Personals
Tuan Hue, THI (huetuan1984@gmail.com) / Updated June 2007 || View My Stats