I don't think you can just "test" KinFu for that purpose, it is much more work. So I did it and I can share with you experimental code, but even an experienced programmer would need several days to get it working.
What I can offer you:
- Old KinFu Wrapper: the processing is much more concurred. The original PCL-Code is more a prove of concept, very serial and the GPU is idle most of the time
- Projection an occupancy grid on the ground on the GPU and copy the results into the Host memory. Originally, KinFu approximates just the TSDF by a voxel grid (stored on the device memory)
- the approach uses Kinfu large scale, the local map is copied into a dynamic growing global map
- finding the ground (plane coefficients, RANSAC)
- interface which is tested/used with the navigation stack
What is missing:
- rectification of the map
- including odometry of the rebot wheels (It really have to be done)
- Kalman filter or similar
- adjust to your robot
- wlan optimization, due to the data traffic
- solving problems due to the lack of loop closure capability of the PCL implementation
See also: https://answers.ros.org/question/142461/pcl-kinfu-in-ros-hydro/