Ask Your Question

Soft Cascade - sorting detection results

asked 2013-05-14 04:03:15 -0500

luko28 gravatar image

Hi, I'm using Soft Cascade from OpenCV 2.4.9 for detecting pedestrians in video. Detector is working quite well, but I have a little problem with results. Detector is returning quite lot of bounding rectangles of one person for almost every detection. Example: there are three pedestrians and ~30 bounding boxes for them, ~10 for person... The problem is bigger when people are walking in groups, then I just get a one big cloud of rectangles... Setting confidence threshold for results is probably not the solution, lot of results has quite nice lvl of confidence. So the question is, how to sort detection results to get just best detections per object. Something similar like "CV_HAAR_FIND_BIGGEST_OBJECT" in Haar detection function, that would work also when detected objects are close together. Anybody have any ideas?

edit retag flag offensive close merge delete


Hi, Did you train the softcascade by yourself. How to write the softcascade training config file. could you give me some advices

samuel_1208 gravatar imagesamuel_1208 ( 2013-05-15 03:28:32 -0500 )edit

1 answer

Sort by ยป oldest newest most voted

answered 2013-05-14 04:43:17 -0500

updated 2013-05-14 04:57:11 -0500

Did you try the Group Rectangles functionality proposed here? It's seem to be your issue.

edit flag offensive delete link more


Well I didn't know about that function, after short tests, I see that it is something that i was looking for :) Thanks for help! Also, I didn't saw before that there is something like "rejection criteria" in Soft Cascade that also helps with this kind of problems. By default it is disabled.

luko28 gravatar imageluko28 ( 2013-05-14 06:07:34 -0500 )edit
Login/Signup to Answer

Question Tools


Asked: 2013-05-14 04:03:15 -0500

Seen: 298 times

Last updated: May 14 '13