Geek asks and answers

Proposal for an object recognition algorithm using depth information

I recently implemented recognition software, following the methodology described in this article . However, my dataset also contains depth maps made using OpenNI.

I would like to increase recognizer reliability using depth information. At least about training 1-vs-all SVMs computes histograms of the response from the response after extracting the VFH descriptors (I adapted the OpenCV DescriptorExtractor interface for this task). But the point is, how can I combine two things to get more accurate results? Can someone suggest me a strategy for this?

Ps I would very much like to test a recognizer that directly shows objects in kinect (and not the way I do now by feeding cropped images to the recognizer).

+5
source share
1 answer

I suggest you take a look at PCL, which is a structure similar to opencv, only it is designed to handle cloud computing. Some time has passed since it was used, but algorithms are other modern versions.

0
source

More articles:


All Articles