Ask Your Question

Accuracy of face detection with haarcascade

asked 2013-03-16 12:16:44 -0500

Leo gravatar image

Hi, Recently I am confused by some strange things occur when I do face detection with "haarcascade_frontalface_alt2.xml"

Say when detecting my face in the video captured by web camera, although neither my face nor the camera moves, the coordinates and size of face returned would change a little. Why is this going to happen? Is there a way to improve it?


edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted

answered 2013-03-17 14:16:17 -0500

This is because you are basically using a cascaded structure for each frame. The input data from your webcam has slight lightning variances in each frame, leading to a possible different detection result. Since the algorithm is combining overlapping detection windows, and a single frame could lead to another amount of overlapping frames, it is possible to have small changes.

If you want it more constant, then apply an averaging filter over frames, fitting a fixed size box to the center of the output.


edit flag offensive delete link more


Thank you Steven. It seems I need someone to take over this work...

Leo gravatar imageLeo ( 2013-03-17 21:43:05 -0500 )edit

Or you can try some suggestions and report with the problems. I will gladly help out where possible, so will others on this forum, if you show that you put in some effort.

StevenPuttemans gravatar imageStevenPuttemans ( 2013-03-18 08:10:42 -0500 )edit

Thanks. I'll try it.

Leo gravatar imageLeo ( 2013-03-18 14:04:23 -0500 )edit

Question Tools


Asked: 2013-03-16 12:16:44 -0500

Seen: 657 times

Last updated: Mar 17 '13