RPG MONOCULAR POSE ESTIMATION TUTORIAL
OVERVIEW
The goal of this project was to develop a tutorial for using the RPG Monocular Pose Estimator ROS package that was developed based on the paper cited below. When used properly, the monocular pose estimator package can prove to be extremely useful in applications where detecting small changes in position accurately is essential.
M. Faessler, E. Mueggler, K. Schwabe, D. Scaramuzza: A Monocular Pose Estimation System based on Infrared LEDs. IEEE International Conference on Robotics and Automation (ICRA), Hong Kong, 2014.
As an extension, we wanted to be able to detect changes in position of a device with IR LEDs mounted on it and use that data to control the movement of a turtle in the turtlesim package.
This was one of the projects in our ME495 course in the MSR program. My partner was ChuChu Igbokwe.
MAJOR STEPS INVOLVED
Camera calibration using the appropriate ROS package
Following the documentation for the RPG Monocular Pose Estimator package to try and getting the package running
Building a 3D holder for your LEDs and calculating the transforms between them
Building a circuit for powering the LEDs
Fine-tuning parameters of the tracking algorithm
Building a ROS package to run a demo involving the control of the turtlesim using this package
The link to our final tutorial explaining all of these steps in detail is here: https://github.com/ChuChuIgbokwe/RPG-Monocular-Pose-Estimation-ME495-HW2
FUTURE STEPS
We would ultimately like to take this project to one step beyond this and see how different variables in the system (external as well as factors relating to software) affect the image with detections output. For example, want to test the difference in the output images with detections and publish rate of the /monocular_pose_estimator/image_with_detections node as well as whether different filters and different LEDs would impact the output we are getting positively or not. We still believe that there is a reason for the bottleneck of information that is happening between the usb_cam and monocular_pose_estimator node and would like to spend more time on figuring out why that is.