Object Tracking measure

This project aims to calculate metrics for tracking algorithm (especially MOTA, IDF1)

MOTA

See [1].

const otm = require('object-tracking-measure');

const groundTruths = [
    [
        [22, 33, 20, 20],// x, y, w, h
        [22, 33, 20, 20],
        [22, 33, 20, 20],
        [22, 33, 20, 20]
    ],
    [
        [22, 33, 20, 20],// x, y, w, h
        null,
        [25, 35, 20, 20],
        [39, 41, 20, 20]
    ]
];

const predictions = [
    [
        [23, 33, 22, 20],// x, y, w, h
        [21, 35, 20, 26],
        [23, 33, 22, 20],
        [21, 35, 20, 26]
    ],
    [
        [23, 33, 20, 20],// x, y, w, h
        null,
        [23, 35, 22, 20],
        [39, 35, 20, 26]
    ]
];

otm.mota({
    groundTruths,
    predictions
});

IDF1

See [2].

const otm = require('object-tracking-measure');

const groundTruths = [
    [
        [22, 33, 20, 20],// x, y, w, h
        [22, 33, 20, 20],
        [22, 33, 20, 20],
        [22, 33, 20, 20]
    ],
    [
        [22, 33, 20, 20],// x, y, w, h
        null,
        [25, 35, 20, 20],
        [39, 41, 20, 20]
    ]
];

const predictions = [
    [
        [23, 33, 22, 20],// x, y, w, h
        [21, 35, 20, 26],
        [23, 33, 22, 20],
        [21, 35, 20, 26]
    ],
    [
        [23, 33, 20, 20],// x, y, w, h
        null,
        [23, 35, 22, 20],
        [39, 35, 20, 26]
    ]
];

otm.idf1({
    groundTruths,
    predictions
});

Advanced usage

By default, object-tracking-measure uses

distance between boxes is (1 - Intersection Over Union) (using mean-average-precision library)
threshold is 1 (i.e. IOU = 0 - no overlap)

You can cutomize this, for example to track distance between {x,y} points like

const otm = require('object-tracking-measure');

const groundTruths = [
    [
        {x: 22, y: 34},
        {x: 22, y: 34},
        {x: 22, y: 34},
        {x: 22, y: 34}
    ],
    [
        {x: 55, y: 68},// x, y, w, h
        null,
        {x: 55, y: 68},
        {x: 55, y: 68}
    ]
];

const predictions = [
    [
        {x: 22, y: 34},// x, y, w, h
        {x: 22, y: 34},
        {x: 22, y: 34},
        {x: 22, y: 34}
    ],
    [
        {x: 55, y: 68},// x, y, w, h
        null,
        {x: 55, y: 68},
        {x: 55, y: 68}
    ]
];

otm.idf1({
    groundTruths,
    predictions,
    distFn: ((a,b) => Math.sqrt(((a.x - b.x) * (a.x - b.x)) + ((a.y - b.y) * (a.y - b.y)))), // Euclidian distance
    threshold: 2 // means that 2 meters far is too far
});

Inspect ID Metric

const measure = otm.idDetails({
    groundTruths,
    predictions
});

console.log(otm.idInspect(Object.assign({}, measure, {
    columns: process.stdout.columns - 20
})))

will print

--
GroundTruth[0]✓――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――✓
Prediction[0] ✓――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――✓
              |----------------------------|----------------------------|---------------------------
              0                            1                            2                           

--
GroundTruth[1]✓―――――――――――――――――――――――――――✓?―――――――――――――――――――――――――――?✓――――――――――――――――――――――――――✓
Prediction[1] ✓―――――――――――――――――――――――――――✓?―――――――――――――――――――――――――――?✓――――――――――――――――――――――――――✓
              |----------------------------|----------------------------|---------------------------

Inspect MOT Metric

const measure = otm.motDetails({
    groundTruths,
    predictions
});

console.log(otm.motInspect(Object.assign({}, measure, {
    columns: process.stdout.columns - 20
})))

will print

0[0]                1-1-1-1-1-1-1-1-1-1-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-
1[1]                0-0-0-0-0-0-0-0-0-0---------------------1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-

Other tools

getStats

const result = otm.getStats({
    track: [
        [22, 33, 20, 20], // X, y, w, h
        null,
        [25, 35, 20, 20],
        [39, 41, 20, 20],
        null
    ]
});

/* 
{
    count: 3, // number of non-null point)
    iterationAge: 1, // number of null at the end
    fullDensity: 0.6, // non-null /total size of track
    gapDensity: 0.3333333333333333, // number of gaps / number of non-null
    density: 0.75, // non-null / size of the trimed track
    firstIndex: 0, // first index of the trimed track
    lastIndex: 3 // last index of the trimed track
}
*/

fastGetNullSegment

const result = otm.fastGetNullSegment({
    track: [
        [22, 33, 20, 20], // X, y, w, h
        null,
        null,
        null,
        [25, 35, 20, 20],
        [39, 41, 20, 20],
        null
    ]
});

/* 
{
    first: 1,
    last: 5,
    type: 'null',
}
*/

References

[1] Keni Bernardin and Rainer Stiefelhagen (2008). Evaluating Multiple Object Tracking Performance: The CLEAR MOT Metrics

[2] Ergys Ristani1, Francesco Solera2, Roger S. Zou1, Rita Cucchiara2, and Carlo Tomasi1 (2016). Performance Measures and a Data Set for Multi-Target, Multi-Camera Tracking

Package detail

object-tracking-measure

readme