Orientation Tracking and Panorama Generation

In this project, we implement a quaternion based orientation estimation and panorama generation using IMU and camera data. We optimize these estimates using PGD on the measurements from the IMU. These are then used to generate a stitched panorama image.

Figure 1: Panorama image generated from stitched camera frames using optimized orientation estimates.

Installation

Follow the steps to setup the environment and install packages required to run this project.

conda create -n <ENV_NAME> python=3.8
conda activate <ENV_NAME>
pip install -r requirements.txt
pip install --upgrade "jax[cpu]"

Folder Structure

The project is organized as follows:

.
├── data
│   ├── cam
│   ├── imu
│   └── vicon
├── img
├── utils.py
├── model.py
├── optimizer.py
├── panorama.py
├── quaternion.py
└── main.py

Running the code

Use the following command to run the code.

python main.py --dataset <DATASET_NO>

Arguments:

• --dataset <DATASET_NO>: Select any dataset from 1-11.
• --alpha <ALPHA> (Optional ) : Learning rate for the optimization (default: 0.001)
• --max_iter <MAX_ITER> (Optional ) : Maximum iterations for the optimization (default: 100)

Example:

python main.py --dataset 10 --alpha 0.001 --max_iter 80