David's Website

LiDAR Point Cloud from Sparse to Dense using GAN (2019/09)

I build a GAN model based on Pix2Pix to learn how to transform the point cloud projection image into a camera depth image.

With camera matrix and the static transform between LiDAR and camera, I can easily transform the coordinate from point cloud to image, and vise versa.

The code is available at my github repo: https://github.com/championway/lidar_S2D

SubT Challenge NCTU Introduction (2019/08)

The DARPA SubT Challenge is a competition which seeks novel approaches to autonomous rapidly map, navigate, and search artifacts in an unknown underground environments during time-sensitive combat operations.

Team NCTU represent Taiwan to go for this competition. This is also not only ours but also Taiwan's first team who participate in DARPA competition.

We got 7th place in this competition.

Marine Robot Tracking with Image and Point Cloud (2019/03)

Marine robot tracking and following with image and point cloud.

I use SSD (Single Shot Multibox Detector) and kalman filter to track the front robot, and use PID controller to follow the robot.

I use point cloud to track two marine robots by Point Cloud Library (PCL), and then visualize the trajectories on Rviz.

Multi-view CNN Point CLoud Classification (2019/01)

The goal is to classify point cloud data from LIDAR sensor, and I use Point Cloud Library and deep learning to implement.

This work is focus on using deep learning to classify 3D point cloud with 2D point cloud projection image.

I also use this method in RobotX 2018 real-world competition, and got a good performance.

Maritime Robotx Challenge (2018/12)

The challenge is about developing and using autonomous maritime systems on autonomous surface vehicle platforms and sensors held in December 2018 in Hawaii.

3D perception leader, I am in charge of 3D LIDAR and depth camera perception, including detection and classification using deep neurual network.

We, team NCTU, got 5th place in RobotX competition, and this is also the first time that we participated in RobotX competition.

website / pdf

NCTU self-driving robot (2018/06)

This project is to build a robot which can do self-navigate tast in campus with dynamic obstacle.

The robot has IMU, GPS, wheel odometry and a 16-bin LIDAR on it.

pdf

3D Object Modeling (2018/01)

Nowadays, people are using 3D information more often to do their work than 2D visual information. As a result, I use PCL, ICP, iSAM and apriltags to build a 3D point cloud modeling system with only one depth RGB camera

pdf

Multi-robot patrolling system (2017/12)

We realize that robot can help us with something routine and cycling, just like patrolling, which we should patrol lots of nodes again and again. As a result, I build a multi-robot patrolling system using ROS. It can apply to surveillance & security, and production line delivery.

pdf