Multi-instance semantic segmentation of point clouds

Project's contributors:

 

Project's goals:

  • Mounting sensors (camera and LiDAR) on our vehicle.
  • Sensor calibration
  • Data acquisition
  • Multi-instance semantic segmentation of camera data
  • Back-projection of multi-instance semantic segmentation onto the LiDAR's points of cloud.
  • Removal of instances of dynamic objects.
  • Generation of SLAM (without dynamic objects).

 

Project description:

I've start with studying the EfficientPS Deep Neural Network to understand how it works and if it could be good for get the multi-instances semantic segmentation of the camera's images for the project.

Work still in progress ...

 

Check out other projects about robotic perception.