Welcome

Welcome to my research homepage. I'd like to share through this platform my research work and ideas.

Kai Li

PhD student, Data System Group, University of Central Florida

I am currently a Ph.D. Student in Department of Electrical Engineering and Computer Science at University of Central Florida (UCF) , advised by Dr. Kien A. Hua. I am a member of Database Systems Group (DSG) at UCF. In 2011, I was honored enough to be admitted to UCF with the highest honor for new graduate students-the Trustee Doctorl Fellowship. Thanks to the sufficient financial support from this fellowship, I have been able to focus on my research and quickly tranform myself from a coursework-oriented undergradute into a mature researcher. My current research interests include multimedia data analysis, machine learning, computer vision/graphics. I have also published some works on the topic of wireless sensor networks. Before coming to UCF, I got my B.E. degree majoring in Automatic Control from Huazhong University of Science and Technology (HUST) , Wuhan, China in 2010.

Education

  • University of Central Florida
    2011 - present

    Ph.D in Computer Science, Department of Electrical Engineering and Computer Science.

    4000 Central Florida Blvd., Orlando, FL 32816, USA

  • Huazhong University of Science and Technology
    2006 - 2010

    B.E. in Automatic Control, Department of Control Science and Engineering

    1037 Luoyu Rd., Wuhan, Hubei 430074, China

Publications

  • Cross-modal Hashing Through Ranking Subspace Learning
    [pdf] [slides] [code]
    Kai Li, Guojun Qi, Jun Ye and Kien A. Hua
    IEEE International Conference on Multimedia and EXPO (IEEE ICME) , July 2016. (Oral acceptance rate 15% )
  • WTA Hash-based Multimodal Feature Fusion for 3D Human Action Recognition
    Jun Ye, Kai Li and Kien A. Hua
    IEEE International Symposium on Multimedia (IEEE ISM) , December. 2015. (Best Paper Award)
  • Temporal Order-Preserving Dynamic Quantization for Human Action Recognition from Multimodal Sensor Streams
    [pdf]
    Jun Ye, Kai Li, Guojun Qi and Kien A. Hua
    ACM International Conference on Multimedia Retrieval (ACM ICMR) , June. 2015.
  • ThingStore: A Platform for Internet-of-Things Application Development and Deployment
    [pdf]
    Kutalmis Akpinar, Kien A. Hua and Kai Li
    ACM International Conference on Distributed Even-based Systems (ACM DEBS) , June. 2015.
  • What's Making that Sound ?
    [pdf] [slides]
    Kai Li, Jun Ye and Kien A. Hua
    ACM Multimedia Conference (ACM MM) , Dec. 2014.
  • Mobility-assisted Distributed Sensor Clustering for Wireless Sensor Networks
    [pdf] [slides]
    Kai Li and Kien A. Hua
    IEEE Global Communications Conference (GLOBECOM) , Dec. 2013.
  • Mobile Data Collection Networks for Wireless Sensor Networks
    [pdf] [slides]
    Kai Li and Kien A. Hua
    International Conference on Multimedia Communications, Services and Security (MCSS) , Jun. 2012.

Professional Activities

  • Software Engineering Intern with HP Vertica
    May 2015 - Aug. 2015

    I worked on the massively-distributed and highly-parallel columnar database system. Specifically, I applied robust optimization theory to database physical design. I implemented a robust physical design algorithm and integrate it with Vertica's existing DBDesigner. The algorithm is able to maintain stable performance against perturbations in query workloads.

  • Student Volunteer of IEEE Infocom 2012.
    Mar. 2012

    I served to hand out badges, programs, and answer meeting-related questions from arriving registrants. I also assisted with setting up of the conference room, directing attendants towards the right program, and collecting and delivering evluation forms at the begining and end of the paper presentation.

  • Research Intern with Institute for Pattern Reconigtion and Artificial Intelligence, HUST.
    Sep. 2010 - Apr. 2011

    Reserach on automatic object recognition using template matching algorithms. We built 3D models of the scene and obtained sequences of object templates with different scales and views from simulations. Those templates were then used to find matches in videos of real scene to identify the desired target. We statistically evaluated the performance of different templates by varying the simulation environment and formulated principles for optimal selection of templates.

  • Research Intern with National Lab of Pattern Recognition, CASIA.
    Jul. 2009 ~ Aug. 2009

    I worked on an object classification research project during this internship. The object classfication algorithm we used is an emulation of the process of object recognition in primates' visual cortex. A set of scale- and position- tolerant features were constructed to approximate the response properties of cells along the ventral stream of visual cortex. This work was accepted for publication in Proc. of CVPR. As a junoir undergraduate with only course work experience back then, I was able to grasp this new research topic in a very short time thanks to my strong motivation and solid academic background. I further devloped a system prototype with the object classification algorithm as the core. This system provides intuitive interface for visulizating and tweak various aspects of the algorithm.

Awards and Honors

It is my great honor to have my hard work and achievements recognized by my universities and departments. Below are some important awards I have won in the past few years.


  • 2015. 2nd Runner-up for Best Intern Poster Presentation at HP Vertica
  • 2013. Graudate Presentatino Fellowship
  • 2011~2014. Trustee Doctoral Fellowship (the top scholarship for UCF graduate students with scholarships and tuition waivers for four years)
  • 2010. Excellent Graduate Award (top 10%)
  • 2009. National Scholarship for Encouragement (top 3%)
  • 2008. National Scholarship (top 2%)
  • 2008. Foxconn Scholarship (top 5%)
  • 2007. Excellent Award of C Language Programming Competition (top 3)

Projects

Go to next/previous page

Complex Event Processing of Multimodal Information Flow for Internet of Things

An advanced app-store concept, called ThingStore, is introduced in this paper. It provides a "market place" environment to facilitate collaboration on Internet-of-Things (IoT) applications development, and a platform to host their deployment. ThingStore services three categories of users: Thing Provider: "Things" (such as online cameras and sensors) can be made more intelligent through event detection software routines called smart services. A thing provider may deploy "things" and advertise their smart services at ThingStore; Software Developer: Software developers can develop apps that query relevant smart services using EQL (Event Query Language) much like traditional database applications can be conveniently developed atop a standard database management system today; End Users: An end use may subscribe to a particular app for event notification. In this IoT architecture, ThingStore is a computation hub that links together human, "things", and computer software in a cyberphysical lifecycle to enable fusion of human and machine intelligence to accomplish some common goal. Not only human, but also "things", may adjust the physical world. New changes in the physical world may, in turn, incur new event detections and therefore initiate another cycle of this ecology-inspired computational lifecycle.


Relevent Publications

  • ThingStore: A Platform for Internet-of-Things Application Development and Deployment
    [pdf]
    Kutalmis Akpinar, Kien A. Hua and Kai Li
    ACM International Conference on Distributed Even-based Systems (ACM DEBS) , June. 2015.

Multimodal Feature Fusion for Real-time Smoke Detection in Videos

We identified several potentially useful features in different papers and combine them in our smoke detection algorithm. Our smoke detection algorithm involves four major steps: background modeling, candidate region extraction, feature extraction, and detection. We use a decision-fusion method to combine multimodal features in the same classification framework. We implemented the algorithm in C++ using OpenCV libraries and achieved promising results.


Multimodal Information Correlation Analysis

Mutimodal information fusion has shown to be promising for many tasks that used to exploit information from a single modality. An intuitive explanation to this fact is that real-life events are inherently multimodal, while an unimodal system utilizes only part of the information leaving the rest wasted. One particular research area that takes advantage of multimodal information fusion is audiovisual analysis that combines audio and visual information. The combination of audio and visual information, however, should not be totally random. But rather, only correlated information are combined for improved performance. The primary goal of this work is to find correspondences between audio and visual modalities through correlation analysis. In this work, I first use a novel video segmentation algorithm to decompose the entire video into a number of spatial-temporal region tracks and represent each track as the time series of a motion feature. Audio signals are represented using the time series of audio energy. Audio and visual representions are further embedded in a common ordinal space using Winner-take-all hashing. The audiovisual correlation is thereby transformed into a problem of Hamming distance computation of the resultant rank space embeddings. The proposed approach has shown superior performance over state-of-the-art.


Relevent Publications

  • What's Making that Sound ?[pdf]
    Kai Li, Jun Ye and Kien A. Hua
    ACM Multimedia Conference (ACM MM) , Dec. 2014.

Multimodal Biometric Authentication

Biometric authentication is an emerging technology that utilize biometric data for the purpose of person identification or recognition in security applications. Among the widely used biometrics, voice and face traits are most promising for pervasive application in everyday life, because they can be easily obtained using unobtrusive and user-friendly procedures. For quite a long time, the use of acoustic information alone has been a great success for speaker authentication applications. Meanwhile, the last decades or two also witnessed great advancement in face recognition technologies. However, in adverse operating environments, neither of the two techniques achieves optimal performance. Since visual and audio information conveys correlated and complimentary information to each other, integration of them into one authentication system can potentially increase the system’s performance, especially in suboptimal operating conditions. In this project, I made an extensive survey on state-of-the-art authentication technologies based on the fusion of audio and visual biometrics. In particular, various aspects of different existing biometric systems are analysed and compared. In addition, a novel idea of dynamic 3D audio-visual biometric authentication exploiting the Microsoft Kinect is presented.


Relevent reports

  • Identity Authentication based on Audio Visual Biometrics: A Survey[pdf][slides]
    Kai Li

Mobile Data Collection Network for Wireless Sensor Networks

Optimizing energy consumption and improving the lifetime of micro-sensors have been a heated research topic in the area of wireless sensor networks (WSN). Energy-effient solutions have been sought in various aspects of WSNs, such as routing, medium access control, duty cycle scheduling, etc. In this project, we address the energy consumption problem in sensor data collection. Previous research have mostly explored all sorts of clustering protocols with different cluster head selection criteria and the performance has been pushed to the limit with the current paradigm. Motivated by wireless connected robots, we consider a novel approach to WSN design, in which a separate autonomous mobile ad hoc network (aMANET) is used for data collection. An aMANET is different from a traditional MANET in that an aMANET node can move autonomously and cooperatively to accomplish some common task. In our environment, this task is to collect sensor data and forward them to the sink. We designed several decentralized protocols and explored the performance gain through both mathematical modeling and network simulation. Our study results showed that the addition of mobile elements opens up new possibilities to optimize sensor energy consumption to the next level.

Relevent Publications

  • Mobility-assisted Distributed Sensor Clustering for Wireless Sensor Networks [pdf][slides]
    Kai Li and Kien A. Hua
    IEEE Global Communications Conference (GLOBECOM) , Dec. 2013.
  • Mobile Data Collection Networks for Wireless Sensor Networks [pdf][slides]
    Kai Li and Kien A. Hua
    International Conference on Multimedia Communications, Services and Security (MCSS) , Jun. 2012.

Resume/CV

Go to next/previous page

Miscellaneous

(Under Construction ...)

Go to next/previous page

I love photography in my spare time, because I like to freeze the stunning beauty and amazing moments in my life. One doesn't have to be professional to get into photography. It's all about you and your life. Here are some of the photos I have taken over the past years.

  • All
  • Landscape
  • Animals
  • People
  • Seasons

Contact Me

Go to next/previous page

Contact info

If there're any questions about my work, please contact me through email. I'm more than happy to discuss about any ideas related to my research.

  • 4000 Central Florida Blvd., Orlando, FL 32826, USA

  • kaili AT eecs DOT ucf DOT edu
  • kailee DOT ucf AT gmail DOT com (Prefered)