ocam

ocam: omnidirectional project points

A torch, numpy, MATLAB for python, and python implementation of the projection functions of Scaramuzza's Omnidirectional Camera.

Installation

Clone the repository and install

git clone https://github.com/Chris10M/ocamcalib_python.git
cd ocamcalib_python
python -m pip install .                  

Optional Dependacies

The conversion scripts for MATLAB intrinsics to JSON intrinsics and JSON to MATLAB intrinsics require MATLAB and MATLAB Engine API for Python.

Tested with MATLAB R2023a

Usage

Command Line

convert_intrinsics_to_matlab_fisheye path/to/fisheye_intrincis.json # output: path/to/fisheye_intrincis.mat
convert_matlab_to_intrincis_fisheye path/to/fisheye_intrincis.mat # output: path/to/fisheye_intrincis.json

The fisheye_intrincis.mat can be obtained by following the instructions here or using the Camera Calibration App. I would recommend going with the latter as it is very easy.

Minimal usage script

import ocam
import numpy as np

# Specify the path to the camera calibration JSON file
ocam_calib_json = 'intrinsics.json'

# Load the camera calibration model from the JSON file
ocam_calib_model = ocam.to_ocam_model(ocam_calib_json)

# Generate random 3D points (replace this with your actual 3D points)
num_points = 10  # Number of 3D points
points3D = np.random.rand(num_points, 1, 3)  # Random 3D points with x, y, z coordinates - Nx1x3

# Project the 3D points into 2D points using the camera calibration parameters
points2D = ocam.world2cam_np(points3D, ocam_calib_model) # points2D - Nx1x2

# Print the resulting 2D points
print("Projected 2D points:")
print(points2D)

The intrinsics.json has the following format,

{
    "pol": [161.14670546203945, 0, -0.00258314253679925, 6.035124622290237e-06, -2.5573236046748687e-08],
    "width": 640,
    "height": 480,
    "xc": 321.3651091048847,
    "yc": 236.25516320600983,
    "c": 1.0,
    "d": 0.0,
    "e": 0.0,
    "shift_cx": 1.5283916846915417,
    "shift_cy": -4.035136165139761,
    "invpol": [246.60052640819538, -144.4605008632319, -6.62349373938239, -18.73714768763381, 14.831092227950778, 2.7995346430010786, 0.22537325854641585, -4.961875735767433, 1.7473552065804472]
}

the invpol in computed in ocam.to_ocam_model function.

More examples:

Check out example.py for more examples.

Citations

If you find this code useful for research, consider citing:

• Original Omnidirectional camera publication,

@article{scaramuzza2014omnidirectional,
  title={Omnidirectional camera},
  author={Scaramuzza, Davide and Ikeuchi, Katsushi},
  year={2014},
  publisher={Springer US}
}