detection of people [closed]

asked 2015-04-30 12:53:41 -0500

jorge gravatar image

hello and I been working with haar opencv detector to detect people but I not brought very good results detected, which may be the problem training or detection method for people haar nis is very good appreciate your answers

edit retag flag offensive reopen merge delete

Closed for the following reason question is not relevant or outdated by sturkmen
close date 2020-12-09 03:25:39.565321


Simply said, use google... opencv cascade classification for pedestrians is known to have poor results, since the algorithm is actually not suited for pedestrian detection. Go to HOG+SVM, LatentSVM, ... approaches, which are way more suitable!

StevenPuttemans gravatar imageStevenPuttemans ( 2015-05-04 04:15:11 -0500 )edit

Hi thanks for answering. knowing that it is wrong to'm working method cabesa recognizing only people at the entrance and exit of a door with the camera located at the top for counting.

the recognition is poor but is achieved, what I want to achieve now would overcome the occlusion with two cameras through haar.

in other words I utiliso haar to detect pesonas in both houses but how people detect both houses making him saver that is the same person and one count delete.?

jorge gravatar imagejorge ( 2015-05-05 10:52:52 -0500 )edit

"with the camera located at the top" - this may well be the reason for your poor detection. iirc, opencv's face cascades are trained on profile / front facing views. if you're using the hogcascade_pedestrians.xml, - again, trained on front/profile views, not from top-down.

berak gravatar imageberak ( 2015-05-05 11:04:08 -0500 )edit