Bachelor Topics 2019

 

For ShanghaiTech Bachelor students graduating in 2019 the MARS group is offering the following topics. You will be co-advised by one of my graduate students.

1. 3D Camera Field of View Webtool

Topic:

The goal of this project is to write a tool to help vision researchers to plan the arrangement of their cameras, with respect to their field of view.
You will develop a 3D visualisation tool based on web technology (JavaScript), so it can run directly in the browser. The tool will visualize the field of view of multiple cameras in 3D:  Show the Frustum of the but also show a sphere (with different radii) with the observed areas highlighed (also such that overlapping areas are visible). Maybe come up with other good visualisations. The project will also will need data input forms for adding cameras and their configuration (field of view, pose (position and orientation)).

Ideally the project will progress quickly so you can also work on an extension to optimally arrnage a number of cameras based on some criterion.

We will use JavaScript and one of the JavaScript 3D engines: https://en.wikipedia.org/wiki/List_of_3D_graphics_libraries#JavaScript-based_engines.

 Prerequisites:
  • Hard requirement: CS student
  • Hard requirement: Experience with 3D programming (WebGL, OpenGL, Direct3D, Metal, Vulcan, OGRE, ...)
  • Hard requirement: Experience with web programming: JavaScript (AJAX, Node.io, ...)
  • Big plus: Experience with Javascript 3D engines.

I will only accept a student that fullfills all hard requirements.

Graduate Co-Supervisor:  Haofei Kuang

2. Fast 3D PointCloud Coloring

Topic:

The MARS Lab has a mapping robot equipped with 2 advanced 3D Laser Scanners (Velodyne 32) and 9 high resolution cameras (5MP). Also, there is a similar setup for a sensor rig that is being put on a car for research on large area mapping and autonomous driving. We have working code to assign RGB color values to the laser beams by using the camera images - but it is slow and not optimized for multiple cameras. Your task will be to improve the speed of the point cloud coloring (we know how in principle, but you can also find additional tricks). If you are fast you can also work on optimizing the results (e.g. filter noise points) or on 3D mapping afterwards.

Prerequisites:
  • Hard requirement: CS student
  • Hard requirement: Experience with C++
  • Hard requirement: Experience with multi-threading (OpenMP)
  • Big plus: Experience with 3D point clouds and/ or computer vision
  • Plus: Experience with ROS

I will only accept a student that fullfills all hard requirements.

Graduate Co-Supervisor: Hongyu Chen 

3. Slip Detection via Learning

Topic:

The MARS Lab has a small tracked (RoboCup) rescue robot. Tracked robots often slip (the motors and tracks turn, but the robot is not diving accordingly), especially when turning or on difficult terrain. When the robot is slipping one cannot use the odometry (where did the robot drive based on the motor speeds) for localization and mapping. Especially for monocular visual SLAM odometry is important. 

The goal of this project is to use an IMU sensor (mainly accelerometer) and the motor speeds to constantly compute if the odometry can be trusted or not (so binary output - maybe with confidence level). We think that using some kind of Deep Learning approach would perform well here. You would use the tracking system to capture the ground truth motion of a robot and thus be able to collect lots of training data.

Prerequisites:
  • Hard requirement: Experience with DeepLearning
  • Hard requirement: Experience with Python
  • Big plus: Experience with ROS

I will only accept a student that fullfills all hard requirements.

Graduate Co-Supervisor: Yijun Yuan

4. Differential GPS

Topic:

The MARS Lab has a professional differential GPS system with one base statino and three GPS receivers. The base station has to be in constant communication with the GPS receivers for the system to work. For that it is coupled with a WiFi system. We will install the receivers on the sensor rig for the autonomous car project and on our mapping robots. This thesis has two main work items: a) Learn how to use the GPS system and collect data with it. Also provide some statistics about the accuracy of the GPS system, also compare to other GPS systems (compare to a phone, compare to a "normal" GPS). b) Modify the system such that the information from the base station can also be transmitted via 4G instead of WiFi. 

Prerequisites:
  • Hard requirement: Programming Experience
  • Hard requirement: Experience with Networking (TCP, UDP)
  • Big plus: Experience with GPS
  • Plus: Experience with ROS

I will only accept a student that fullfills all hard requirements.

Graduate Co-Supervisor:  Xiaoling Long

5. IMU RealSense Synchronization

Topic:

The small Rescue Robots use an Intel RealSense D435 camera for SLAM. We are using VINS for SLAM. This software benifits greatly from using an IMU. But for that the IMU and RealSense data have to be time synchronized. Your main task is to use the hardware synchronization techniques both sensors support to collect synchronized data. Further working points can then be the integration of the synchronization system into our existing hardware synchronization solution and to synchronize the two sensors with the tracking system to generate excellent ground truth maps. Alternatively we can also work on internal calibration of the IMU.

 Prerequisites:
  • Hard requirement: Experience with programming (C++, Python)
  • Hard requirement: Experience with circuits (soldering)
  • Big plus: Experience with ROS.

I will only accept a student that fullfills all hard requirements.

Graduate Co-Supervisor:  Zeyong Shan