Ask Your Question

Bounding box ranking / vehicle, pedestrian detection

asked 2017-03-06 11:20:32 -0500

Nbb gravatar image

I have a set of images with hundreds of boxes scattered which aren't shown below. I would like to 'rank' these boxes in terms of how likely it is to contain either a car or a pedestrian similar to what is done in 'Edge Boxes'. The ranked boxes will then be passed to an actual classifier.

Currently, I seem to be getting a lot of boxes around trees and walls. The method needs to be really really fast and so I thought of simply trying to match it with a template that results in the low score on trees / buildings and higher score on cars / pedestrians. I think there is no faster way than direct template matching ?

I was wondering if anyone can provide a suggestion / paper for me to read up on. The focus is on speed.

image description

image description

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted

answered 2017-03-06 18:11:21 -0500

Tetragramm gravatar image

I have an idea. Try using the edge map. Inside each box, calculate a histogram of oriented gradients. The trees have lots of edges in lots of directions. So that's an easy choice. I expect you could separate the pedestrians and cars and buildings fairly similarly.

Normally for identification you do this for small blocks, like THIS. I think you might be able to get away with just one block per bounding box. Worth a shot anyway.

edit flag offensive delete link more

Question Tools

1 follower


Asked: 2017-03-06 11:20:32 -0500

Seen: 507 times

Last updated: Mar 06 '17