Hello all,
I am using opencv for full body detection and wanted to know what would be good settings/limits for the minSize and maxSize parameters? As of now I have them blank, my code seems to detect people fairly well. I also have the minNeighbors parameter set to 7. Each time I run my code I get about 20% false positives. My hope is that setting the size parameters will reduce the amount of false positives.
I do not like the response of @vinnitu simply because ignoring these parameters is plain stupid. The advantage of the minSize and maxSize parameters is that you can limit the range in the object scale pyramid in which instances of your model are being looked for. This means that all false positive detections generated outside those sizes will also be discarded! Therefore it is quite usefull! I am using it in almost every single application where I have a little knowledge about possible object sizes and it increases processing speeds alot!
Imho it it not so important. You can call detectMultiScale with olny 2 first params with default values minSize, maxSize. This do serial resize image and if it possible detect object. I think minSize and maxSize need for fast detect object in more certain circumstances.
Asked: 2015-06-03 16:29:10 -0600
Seen: 679 times
Last updated: Jun 05 '15
