3D obstacle detection and tracking

asked 2019-02-07 15:24:25 -0500

JohannLange gravatar image

Hey folks,

I am looking for a ROS node, that will allow me to detect and track an arbitrary object/obstacle in 3D space. It does not have to detect whether that obstacle is a human, a car, a tree, ... (you get the idea). I am only interested in the current position and the "size" of an obstacle. Approximating the obstacle as a sphere would be enough. Thus, everything the node would need to do is:

  • Detect an arbitrary obstacle in 3D space (point cloud or depth map available)
  • Track that obstacle giving me it's current (relative) position (and velocity)
  • Determining the obstacle's size (spherical approximation is more than enough)

My setup is basically an nVidia Jetson TX2 (Ubuntu 16.04, ROS kinetic) with a stereovision camera, including OpenCV, CUDA, ... Therefore, computation power is widely available.

Thanks folks!

edit retag flag offensive close merge delete

Comments

You're asking for quite a lot there to be honest. Try looking into deep learning approaches for object detection and image segmentation. But I don't know of any ROS nodes close to doing what you want out of the box.

PeteBlackerThe3rd gravatar imagePeteBlackerThe3rd ( 2019-02-07 17:12:43 -0500 )edit

Agreed. I have developed things like that in house for a robotics company because it did not widely exist before and is very dependent on the sensors and situations.

stevemacenski gravatar imagestevemacenski ( 2019-02-07 18:03:28 -0500 )edit

if you find one though, I'd be interested in seeing it and/or contributing some of my elements to it

stevemacenski gravatar imagestevemacenski ( 2019-02-07 18:03:46 -0500 )edit

Well there is a node (or multiple nodes) by Intel for Obstacle Detection and Tracking (IntelROSProject) but they all rely on an x86_64 architecture

JohannLange gravatar imageJohannLange ( 2019-02-08 16:42:40 -0500 )edit