Robust Stereo-Based Person Detection and Tracking can be used in a mobile robotic application.

Color detection and hair detection algorithms.

Dear Redhwan,

I suggest you to see links and attached files on topic.

https://www.researchgate.net/publication/221259161_Vision_Based_Person_Tracking_with_a_Mobile_Robot

https://cecas.clemson.edu/~stb/publications/personfollow_iros2007.pdf

http://jtl.lassonde.yorku.ca/wp-content/uploads/2017/02/pfr_paper_crv2017.pdf

Best regards

Thanks both Sundara Siva Kumar v and Mohamed-Mourad Lafifi

I already use the color feature, it is good and robust. But some issues with illumination change. I need that the robot follows a specific person based on the color and texture of clothes to improve it with illumination change.

Thanks, Aparna Sathya Murthy

YES, I tried it offline and I need to convert it into online (using a camera in real-time). I think that it requires combining with another algorithm for classification.

Apart from GLCM you can use other features : Law's energy measures, Gabor filter, Wavelet transforms

https://openscholarship.wustl.edu/cgi/viewcontent.cgi?article=1812&context=cse_research

https://arxiv.org/ftp/arxiv/papers/1704/1704.05122.pdf

Article Texture classification using wavelet transform

Thanks, Ranjan Parekh

Aparna Sathya Murthy

I used Gabor filter but I can't understand how to classify clothes.

This full my code: https://stackoverflow.com/questions/60932336/how-to-classify-peoples-clothes-by-gabor-filter

Could you help me?

Aparna Sathya Murthy thank you so much

Aparna Sathya Murthy

Sure, I got it but "I can't understand how to classify clothes." as I mentioned above. I am still studying to understand how to classify the clothes of people, I don't know if I need to train data using SVM, for example. Then, use a Gabor filter

Please your opinion if I can use it without training or not?

The output returned by a GLCM or Gabor filter is a matrix not a vector. So the general practice is to calculate some statistical features from the output matrix and put them in a vector. These values should be different for different classes but similar for the same class. Either now you can feed these vectors to a classifier for classification, or even without a classifier you can simply use Euclidean distance to discriminate between the classes. Hope that is clear.

Ranjan Parekh

Thank you a lot a lot for your explanation.

I will go to implement it as you explain.

That's a good question