Star ⭐(Translation only)

Spiral ➰(Translation only)

Arc 🌙 (Translation + Rotation)

This project illustrates a real-time 2D localization algorithm for off-the-shelf mobile phone hardware. We assume that the device is constrained to have only translation and rotation in a 2D space (ex. on a computer desk). We assume that a static planar scene exists at a far enough distance with a decent number of good features to track over time.

The underlying idea is to estimate euclidean transform of features in nth frame with that of 1st frame. As a result,

• When the device is at rest there is translation and rotation readings stay at zero i.e. there is no drift at rest.
• When the video feed is interrupted for some time and then restarted, tracking resumes automatically.
• The speed at which device moves has no effect on tracking as change in pose is only a function of first frame and Nth frame.
• The above point also means that there will be no drift accumulation during the movements.

It is designed to be deployed in a mobile phone with a front/back camera. A simple use case is to use it as an external mouse. A rudimentary inertial odometer using accelerometer with eularian integration is provided as a baseline.