Motivation
This tasks is a part of a project called "Collaborative Robotic Workplace of the Future". The main goal of the project is to develop a workplace where a robotic arm will aid a human worker with an assembly task. A necessary part of the project is the ability to recognize and track the objects (building material, tools) in the workplace.
Brief description of the task
Completing the following sub-tasks will be part of the summer job. Some of the sub-tasks might change or sub-tasks might be added or removed based on your progress. This description should just give you basic overview of the task.
- Compile a list of the state of the art methods for visual object tracking from scientific papers (to make this sub-task easier, there are regular challenges comparing the most current methods, e.g. MOT and VOT)
- Select suitable tracking methods (with the help of the supervisor)
- Test the implementations of the selected algorithms on standardized datasets. Only algorithms with existing source codes or at most easy-to-implement algorithms will be used. Therefore, this sub-task will be mostly about making the existing implementation work and tuning the parameters of the algorithm.
- Test the selected algorithms on a specialized dataset related to the industrial task. Part of this sub-task will be likely the creation of such a dataset (a small, simplified set). A real industrial task is currently substituted with an assembly of a toy building kit (e.g. "Meccano"). Therefore, the set will consist of the building blocks and tools from the kit.
- Compare the performance of these algorithms with "standard" tracking algorithms implemented in the computer vision library OpenCV
Recommended skills
The student(s) will be required to learn these skills in order to accomplish the given tasks. Knowing them beforehand (at least at a basic level) will be advantageous.
- Python
- PyTorch (framework for artificial neural networks in Python)
- OpenCV
- Robot Operating System (ROS)
Notes
Team of two students can work on the task.