Adding support for lidar depth map conversion and attachment to the map

mfehr commented 4 years ago

Adds support for lidar to range image conversions and based on the camera model introduced in aslam_cv PR https://github.com/ethz-asl/aslam_cv2/pull/46
Adds option to convert lidar scans to depth maps in maplab_node. Saves a lot of space.
- Projects to either OpenNI format (uint16_t, [mm]) or Floating point [m]
- Optionally projects color/intensity along with the depth into color/intensity image
Adds option to attach point cloud to map as depth map instead of point cloud.
Add support to have more than one NCamera in the sensor calibration (uses the --selected... flags to determine which one is used for visual mapping). This is requried since in order to attach point clouds as depth maps, we need to add a 3DLidarCamera calibration (=another NCamera) to the calibration file. The same selection/multi-sensor option is implemented for all other sensor types as well.
Some minor fixes and documention for the resource importer tool

Significantly reduces the size of the map when attaching point clouds as depth maps, as can be seen in this example, where the full OS-1 64 beam lidar scans at 10Hz were attached:

9.9G    gonzen_run_7_with_pointclouds
1.5G    gonzen_run_7_with_range_image
2.2G    gonzen_run_7_range_image_and_intensity