Publish Kinect Depth information to a Ros Topic

Question

Hello, all,

Recently, we've been attempting to access the kinect depth data from the published point cloud provided by the openni_node, but we've run into a couple stumbling points:

first, the ros tutorial for pcl_ros

http://www.ros.org/wiki/pcl_ros/Tutorials/CloudToImage

references a topic published by openni called: /camera/depth/points2 but our version of openni_kinect only publishes: /camera/depth/points Are we using an outdated version?

Also, we tried working with /camera/depth/points and we believe that this information is published in binary format. When we attempted to convert the binary format of a single point to x,y,z information though, we received a list of 4 integers. Are we interpreting the information incorrectly?

We also subscribed to the /camera/depth/image, and we could access the depth information from a single point, but we're restricted to meter-by-meter resolution. Is there a better way of getting more resolution with the /camera/depth/image ?

Thanks for reading, and we appreciate any directions you can point us to accessing the depth information (of points) from the kinect.

Sincerely,

Poofjunior

---

Originally posted by Poofjunior on ROS Answers with karma: 88 on 2011-07-13

Post score: 1

Answer 1

The current driver publishes both point clouds and images:

• /camera/depth/points -- this is the non-colored point cloud. To use this, you'd probably want to use PCL (Point Cloud Library) functions to access individual points in a meaningful manner.
• /camera/depth/image -- this is the depth image (not a point cloud). It can easily be converted to an OpenCV image using the cv_bridge package. For using the depth image, it is important to note that is a 32-bit float for each pixel, corresponding to meters. This answer and this one contain some useful information on using the image.

---

Originally posted by fergs with karma: 13902 on 2011-07-13

This answer was ACCEPTED on the original site

Post score: 6

---

Original comments

Comment by fergs on 2011-07-14:
Great! Please click the checkmark next to my answer to accept it.

Comment by Poofjunior on 2011-07-14:
Many thanks, fergs! It turns out that we weren't interpreting the image correctly. We changed cv_image = self.bridge.imgmsg_to_cv(data, "mono8") to cv_image = self.bridge.imgmsg_to_cv(data, "32FC1") and we started seeing much better results.

Comment by Poofjunior on 2011-07-14:
Many thanks, fergs!

Comment by Consider1001 on 2014-04-15:
Thank you for your answer. May I ask a question that if there is a way to get the value(float) of each point on the depth/image_raw before we convert it in to a cvmat?