vican

Very efficient algorithm for camera calibration dockerized

How to use

Step 1 - build the docker image

docker build -t <image-name> .

Step 2 - setup dataset

Create a directory with the following disposal

The folders structure of <object-images> and <cameras-images> is <dataset>/<timestep>/<camera_id>.jpg

<volume-directory>
├── config.txt
├── <object-images>
│   ├── 0
│   │   └── 0.jpg
│   ├── 1
│   │   └── 1.jpg
    ...
│   └── cameras.json
└── <cameras-images>
    ├── 0
    │   ├── 1.jpg
    │   ├── 10.jpg
    │   ├── 11.jpg
    │   ├── 12.jpg
    │   ...
    ├── 1
    │   ├── 1.jpg
    │   ├── 10.jpg
    │   ├── 11.jpg
    │   ├── 15.jpg
    │   ...
    ...
    └── cameras.json

Example of config file - config.txt

object_path:<object-images>
object_calib:cube-calib.pt
cameras_path:<cameras-images>
cameras_pose_est:pose_est.json
aruco:DICT_4X4_1000
marker_size:0.276
marker_ids:0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23
brightness:-150
contrast:120

Example of <object-images>/cameras.json

{
    "0": {
        "fx": 711.1111111111111,
        "fy": 711.1111111111111,
        "cx": 640,
        "cy": 360,
        "resolution_x": 1280,
        "resolution_y": 720,
        "clip_start": 0.10000000149011612,
        "clip_end": 1000,
        "t": [5.816105365753174,21.875133514404297,2.772583246231079],
        "R": [
            [0.03479209542274475,-0.9984201192855835,0.044123295694589615],
            [-0.9954691529273987,-0.038530927151441574,-0.08692909777164459],
            [0.08849187195301056,-0.04089892655611038,-0.995236873626709]
        ],
        "distortion": [0,0,0,0,0,0,0,0,0,0,0,0]
    },
    "1": {
        "fx": 711.1111111111111,
        "fy": 711.1111111111111,
        "cx": 640,
        "cy": 360,
        "resolution_x": 1280,
        "resolution_y": 720,
        "clip_start": 0.10000000149011612,
        "clip_end": 1000,
        "t": [5.816105365753174,21.875133514404297,2.772583246231079],
        "R": [
            [0.03479209542274475,-0.9984201192855835,0.044123295694589615],
            [-0.9954691529273987,-0.038530927151441574,-0.08692909777164459],
            [0.08849187195301056,-0.04089892655611038,-0.995236873626709]
        ],
        "distortion": [0,0,0,0,0,0,0,0,0,0,0,0]
    },
    "2": {
        "fx": 711.1111111111111,
        "fy": 711.1111111111111,
        "cx": 640,
        "cy": 360,
        "resolution_x": 1280,
        "resolution_y": 720,
        "clip_start": 0.10000000149011612,
        "clip_end": 1000,
        "t": [5.816105365753174,21.875133514404297,2.772583246231079],
        "R": [
            [0.03479209542274475,-0.9984201192855835,0.044123295694589615],
            [-0.9954691529273987,-0.038530927151441574,-0.08692909777164459],
            [0.08849187195301056,-0.04089892655611038,-0.995236873626709]
        ],
        "distortion": [0,0,0,0,0,0,0,0,0,0,0,0]
    },
    ...
}

Example of <cameras-images>/cameras.json

{
  "1": {
    "fx": 711.1111111111111,
    "fy": 711.1111111111111,
    "cx": 640,
    "cy": 360,
    "resolution_x": 1280,
    "resolution_y": 720,
    "clip_start": 0.10000000149011612,
    "clip_end": 1000,
    "t": [-4.242201805114746,3.6500678062438965,2.7108936309814453],
    "R": [
      [-0.6672439575195312,-0.35407891869544983,0.6552966237068176],
      [-0.7448351979255676,0.32009637355804443,-0.5854560732841492],
      [-0.0024604161735624075,-0.8787299990653992,-0.4773128032684326]
    ],
    "distortion": [0,0,0,0,0,0,0,0,0,0,0,0]
  },
  "2": {
    "fx": 533.3333333333334,
    "fy": 533.3333333333334,
    "cx": 640,
    "cy": 360,
    "resolution_x": 1280,
    "resolution_y": 720,
    "clip_start": 0.10000000149011612,
    "clip_end": 1000,
    "t": [-1.9186244010925293,3.781113386154175,2.735273838043213],
    "R": [
      [-0.9999966025352478,-0.0022767793852835894,0.001121936016716063],
      [-0.002230584854260087,0.5773668885231018,-0.8164815902709961],
      [0.0012111797695979476,-0.8164814710617065,-0.577370285987854]
    ],
    "distortion": [0,0,0,0,0,0,0,0,0,0,0,0]
  },
  ...
}

Step 3 - calibrate an object

docker run -v ./<volume-dir>:/dataset <image-name> python src/object_calib.py

Step 4 - estimate poses of a camera network

docker run -v ./<volume-dir>:/dataset <image-name> python src/pose_est.py

Auxiliar Scripts

aruco-ids.py

This script finds in every .json the aruco ids and aggregates all ids to complete the config.thx

json file example

{
  "corners": [
    [[[538,574],[487,540],[549,545],[602,579]]],
    [[[460,571],[415,536],[480,540],[535,577]]],
    [[[369,569],[333,532],[407,536],[450,573]]]
  ],
  "ids": [
    [13],[12],[11]
  ]
}

camera-gen.py

To calibrate the object it is needed to assume that the different pictures of the object was taken with different cameras. This script replicates the configuration of one camera to how many you need (in the example the number of cameras needed is N).

cameras.json (before)

{
    "0": {
        "fx": 711.1111111111111,
        "fy": 711.1111111111111,
        "cx": 640,
        "cy": 360,
        "resolution_x": 1280,
        "resolution_y": 720,
        "clip_start": 0.10000000149011612,
        "clip_end": 1000,
        "t": [5.816105365753174,21.875133514404297,2.772583246231079],
        "R": [
            [0.03479209542274475,-0.9984201192855835,0.044123295694589615],
            [-0.9954691529273987,-0.038530927151441574,-0.08692909777164459],
            [0.08849187195301056,-0.04089892655611038,-0.995236873626709]
        ],
        "distortion": [0,0,0,0,0,0,0,0,0,0,0,0]
    }
}

cameras.json (after)

{
    "0": {
        "fx": 711.1111111111111,
        "fy": 711.1111111111111,
        "cx": 640,
        "cy": 360,
        "resolution_x": 1280,
        "resolution_y": 720,
        "clip_start": 0.10000000149011612,
        "clip_end": 1000,
        "t": [5.816105365753174,21.875133514404297,2.772583246231079],
        "R": [
            [0.03479209542274475,-0.9984201192855835,0.044123295694589615],
            [-0.9954691529273987,-0.038530927151441574,-0.08692909777164459],
            [0.08849187195301056,-0.04089892655611038,-0.995236873626709]
        ],
        "distortion": [0,0,0,0,0,0,0,0,0,0,0,0]
    },
    ...
    "N": {
        "fx": 711.1111111111111,
        "fy": 711.1111111111111,
        "cx": 640,
        "cy": 360,
        "resolution_x": 1280,
        "resolution_y": 720,
        "clip_start": 0.10000000149011612,
        "clip_end": 1000,
        "t": [5.816105365753174,21.875133514404297,2.772583246231079],
        "R": [
            [0.03479209542274475,-0.9984201192855835,0.044123295694589615],
            [-0.9954691529273987,-0.038530927151441574,-0.08692909777164459],
            [0.08849187195301056,-0.04089892655611038,-0.995236873626709]
        ],
        "distortion": [0,0,0,0,0,0,0,0,0,0,0,0]
    }
}