Are there any new algorithm about pedestrian detection,people counting?

asked 2013-04-18 14:26:34 -0500

flammxy gravatar image

Are there any new and good algorithm about pedestrian detection,people counting?

And if I use HOG descriptor to detect pedestrian, are there any other tips to improve the result? Because I find my result is not very good.


edit retag flag offensive close merge delete


Hello flammxy, I do the same approach as you do in people counting. My first attempt, using MIT car and people dataset, result in lot of false positives (both test on separate people and car test images). Then I retrain the model using false positives from negative test set. This result in zero detection - please see my post here. After trying on different dataset and modifying parameters (hit threshold, scale, group thereshold), my conclusion is although HOG is so far the most popular people/object descriptor, the result is not perfect (100% hit rate, no false positives). We have to deal with miss and false positives (in my case, 50% hit rate, 50% false positives.

bonchenko gravatar imagebonchenko ( 2013-04-21 21:58:18 -0500 )edit

sorry for double posting, but if you want alternatives to HOG, Latent SVM is based on HOG, and seems to be better at detecting objects. But I cannot find how to create training data for Latent SVM. Anyone know?

bonchenko gravatar imagebonchenko ( 2013-04-21 21:59:35 -0500 )edit

Thank you. But sorry I do not know how to use Latent SVM.

flammxy gravatar imageflammxy ( 2013-04-30 21:59:27 -0500 )edit