Classification using 3D recognition
I am using the 3D recognition http://pointclouds.org/documentation/tutorials/correspondence_grouping.php#correspondence-grouping with Kinect. I want to recognize different objects say, can, bottles and classify them into two classes. For this, I want to use a neural network. I cannot understand how to get the features. How do I extract the features and what will be the features using this code so as to feed into a neural network for classification? Thank you in advance.
Asked by SM on 2014-01-15 07:13:57 UTC
Comments
Hello! Is your question ROS related? ie are you asking how to get feature messages from pcl_ros or are you asking the computer vision foundation behind feature detection?
Asked by Tim Sweet on 2014-01-15 09:29:25 UTC
My question is from the structure "keypoints", how can I access the values of the features?
Asked by SM on 2014-01-16 06:56:15 UTC
Hello! sorry I missed this comment, are you still looking for help? If so, what structure "keypoints" are you referring? Do you have a link to documentation where it is mentioned?
Asked by Tim Sweet on 2014-01-22 17:30:34 UTC
In the link that I gave it just mentions from line 292 that it is computing the keypoints. I am sorry but I do not know where and how they are being stored such that I can use them as raw information into a classifier. I thought that the way data can be extracted from pcl files, in the same way can we extract numeric values from the keypoint structure?
Asked by SM on 2014-01-24 13:35:37 UTC
Oh okay, so this isn't ROS related. I'd suggest using stackoverflow.com for this question, you'll get much more luck finding someone who knows the actual PCL library over there.
Asked by Tim Sweet on 2014-01-24 21:01:35 UTC