Opencv Haar Cascade training / detection for simple objects -
i planning on making cascade detector white cup, red ball, , blue puck. how simple these objects in shape, wondering if there parameter differences have have in training vs finding complex objects such cars / faces? also, within training pos images have objects in different lighting conditions , instances objects under shadow.
for training negative images noticed image sizes may vary. however, positive images must fixed size.
i plan on using 100x100 pos images detect objects 20-30 feet, 200x200 pos images detect objects when within 5ft / directly overhead of object (3 ft off ground appx). mean have train 6 different xmls? 2 each object trained 100x100 , 200x200?
short answer: yes
long answer: probably:
you have think this, classifier going build set of features positive images , use these determine whether detection image same or not. if drastically moving angle of detection, going need different classifier.
let me example pictures:
if @ 20ft away cup looks this:
with associated background/lighting etc, going different classifier if cup looks this(maybe 5ft away different angle):
now, being said, if have larger , smaller versions of cup, may need one. need different classifier each object (cup/ball/puck)
images not mine - taken google
Comments
Post a Comment