Head Image

Zhaoyang Lv  

Research Assistant, Ph.D. in Robotics, Georgia Institute of Technology
Wall Lab & BORG Lab, Institute for Robotics and Intelligent Machines

Previous Education:
M.Sc., Artificial Intelligence in Computing, Imperial College London
B.Sc., Electrical Engineering in Aeronauntics, Northwestern Polytechnical University

I am a 3rd year Ph.D. student in Robotics at Georgia Tech, School of Interactive Computing, jointly advised by Prof. James Rehg, and Prof. Frank Dellaert. I am also working closely with Dr. Zsolt Kira in Georgia Tech Research Institute (GTRI). Before I started my Phd at Gatech, I finished my Master thesis under the supervision of Prof. Andrew Davison at Imperial College London.

I focus on efficient and dense 3D motion (scene flow) estimation from limited amount of images. From a broader perspective, I am interested in general Computer Vision and Robotics problems, especially in exploring perception learning through unuspervised / semi-supervised methods. My strenghts are in the following fields:

  • 3D Scene Flow, Optical Flow and Stereo.
  • Semantic Scene Understanding
  • Structure from Motion, Simultaneous Localization and Mapping

  • Office Address:
    College of Computing Building /
    Robotics & Intelligent Machines Center
    801 Atlantic Drive, Rm. 273B
    Atlanta, Georgia, U.S., 30308

    Email: lvzhaoyang1990 at gmail dot com
    zhaoyang dot lv at gatech dot edu

    Mobile: 404-3458841

    Deep Image Category Discovery using a Transferred Similarity Function

    Yen-Chang Hsu, Zhaoyang Lv, Zsolt Kira


    A Continuous Optimization Approach for Efficient and Accurate Scene Flow

    Zhaoyang Lv, Chris Beall, Pablo F. Alcantarilla, Fuxin Li, Zsolt Kira, Frank Dellaert

    The 14th European Conference on Computer Vision

    We propose a continuous optimization method for solving dense 3D scene flow problems from stereo imagery. As in recent work, we represent the dynamic 3D scene as a collection of rigidly moving planar segments. The scene flow problem then becomes the joint estimation of pixel-to-segment assignment, 3D position, normal vector and rigid motion parameters for each segment, leading to a complex and expensive discrete-continuous optimization problem. In contrast, we propose a purely continuous formulation which can be solved more efficiently. We evaluate our method in the challenging KITTI Scene Flow benchmark, ranking in third position, while being 3 to 30 times faster than the top competitors.

    @inproceedings {Lv16eccv,
    title = {A Continuous Optimization Approach for Efficient and Accurate Scene Flow},
    booktitle = {The 14th European Conference on Computer Vision},
    year = {2016},
    month = {10/2016},
    address = {Amsterdam, The Netherlands},
    author = {Zhaoyang Lv and Chris Beall and Pablo Fern{\'a}ndez Alcantarilla and Fuxin Li and Zsolt Kira and Frank Dellaert}


    Large-Scale Collaborative Semantic Mapping using 3D Structure from Motion Data

    Advisor: Prof. Frank Dellaert Co-advisor: Dr. Zsolt Kira

    A NSF project I am currently working on. My focus is dense 3D scene flow and dynamic 3D mapping.

    KinfuSeg System Image

    KinfuSeg: A Dynamic SLAM Approach Based on KinectFusion
    Master Thesis, Imperial College London, Distinguished Thesis in Department of Computing (3 among 71), Top 5%

    Advisor: Prof. Andrew Davison

    Traditional SLAM methods works under the assumption that the evironment is totally static. When the scene is dynamic, both tracking and mapping will fail. In this project, this system is able to achieve:

    • Tracks the static scene, while segment out the dynamic object.
    • The first solution to real-time fuse dense 3D map for both static and dynamic scenes.

    Bachelor Quadrotor

    Quadrotor Design and its Navigation
    Bachelor Thesis, Northwestern Polytechnical University

    Advisor: Prof. Zhenbao Liu, Prof. Shuhui Bu, Prof. Xiaojun Tang

    The goal of this project is to build up a quad-rotor, with basic navigation and flight control system. The quad-rotor is able to achieve a stable flight with joystick control and hover autonomously.

    Internship at Qualcomm Research, Greater San Diego, May 2016 - Aug. 2016

    Manager: Dr. Ali Agha

    I work in the Sensor Fusion and Motion Planning group. I proposed a factor graph representation for joint Intention Prediction and Motion Planning algorithm. Finally we submitted two patents!
    I also attended a 16-hour hackathon event in Qualcomm using mobile robotics. We proposed a disaster rescue on-site system, using a drone for live video streaming, tracking the mobile vehicles on the road, and sending planning trajectory to guide vehicles out of the disaster. We get in the top-3 out of 50 teams, and present our work to Qualcomm executive members. Great experience both in technical and in business!

    Internship at Zhejiang University, Hangzhou, Dec. 2013 - July 2014

    Mentor: Prof. Guofeng Zhang

    We submit one US patent in indoor real-time robust mapping in extended scale with Huaiwei Research Lab.