Overview
My research interests span the areas of Robotics, Computer Vision, and Artificial Intelligence. I am currently focusing on vision-based robot navigation. More specifically, I develop computer vision algorithms/systems for robots to autonomously navigate in man-made environments like indoor and urban scenarios. In the past, I had research experience in Electrocardiogram (ECG) signal de-noising and pattern analysis.
Research Projects
  • Localization and Mapping for Autonomous Driving
  • NVIDIA's lidar-free autonomous driving is powered by my visual localization work. Check out the demo videos!
  • RGB-D SLAM Using Line Features
  • Large lighting variation challenges all visual odometry methods, even with RGB-D cameras. Line segments are abundant indoors and less sensitive to lighting change than point features. We propose a line segment-based RGB-D indoor odometry algorithm robust to lighting variation. We also investigate fusing point and line features for RGB-D SLAM/odometry. Project Website
  • Multilayer Feature Graph for Robot Navigation
  • We design a multilayer feature graph (MFG) to facilitate scene understanding and robot navigation in urban areas. Nodes of an MFG are features such as SIFT points, line segments, lines, and planes while edges of the graph represent different geometric relationships such as adjacency, parallelism, collinearity, and coplanarity. Project Website
  • Robust Recognition of Planar Mirrored Walls
  • Mobile robots need to recognize objects at their vicinity for navigation and safety purposes. However, highly reflective surfaces, such as glassy building exterior and mirrored walls, challenge almost every type of sensors including laser range finders, sonar arrays, and cameras because light and sound signals simply bounce off the surfaces. Therefore, such surfaces are often invisible to the sensors. Detecting these surfaces is necessary to avoid collisions. In this project, we develop algorithms for detecting planar mirrored walls based on two views from an on-board camera.
  • ECG Signal Processing and Pattern Analysis
  • Vision-based Robotic Pen-and-Ink Drawing
Course Projects
  • CSCE666 Pattern Analysis, 2011 fall
  • Active Learning on Smartphone based Human Activity Recognition
  • In this project, we design a robust human activity recognition system based on a smartphone. The system uses a built-in smartphone accelerometer as the only sensor to collect signals for classification. Active learning algorithms are exploited to reduce the labor and time expense of labeling tremendous data.