Performance Evaluation of Object Tracking Algorithms in Matlab

hi,

How can I evaluate the performance of any tracking algorithm in Matlab. This example of tracking face using KLT and other example using Camshift, how can I compare them?

https://www.mathworks.com/help/vision/examples/face-detection-and-tracking-using-the-klt-algorithm.html

https://www.mathworks.com/help/vision/examples/face-detection-and-tracking-using-camshift.html

Thanks in advance!

Norah

Answer 1

[Drishti]

Hi Nora,

For evaluating the performance of a tracking algorithm, you can follow the below given steps:

• Identifying performance metrics: You can decide a favourable metrics as per the requirement like accuracy, robustness.
• You will need ground truth annotations to serve as a benchmark for evaluating the tracking results.
• Implement the tracking algorithms like KLT and Camshift, execute them both.
• Calculate the metrics, for tracking algorithm you can utilize IoU(Intersection over Union) metrics. Refer to the code snippet below for calculating IoU.

function iou = bboxOverlapRatio(bboxA, bboxB)
    intersectionArea = rectint(bboxA, bboxB);
    areaA = bboxA(3) * bboxA(4);
    areaB = bboxB(3) * bboxB(4);
    unionArea = areaA + areaB - intersectionArea;
    iou = intersectionArea / unionArea;
end

• Compare the IoU value for both the algorithms.

For more information, refer to the MathWorks Documentation of 'rectint' function.

https://www-test1.mathworks.com/help/releases/R2018b/matlab/ref/rectint.html

I hope this helps in getting started.