DROID now processes the list of 2D features which correspond to the static part of the world. It tracks these features in 3D, resulting in a list of 3D world points. Finally, ALTRUISM is fed this list of 3D points and determines which correspond to ``drivable'' or ``road'' points and which correspond to ``obstacle'' or ``off-road'' points. Note that to plan local navigation on an autonomous vehicle it is still necessary to combine the information from ASSET-2 (about the moving objects) with that from ALTRUISM.
Example outputs of the 3D processing and segmentation can be seen in Figures 9 and 10, where successfully tracked 3D points are split into ``road'' and ``off-road'' groups. The moving objects have been ignored and 3D interpretation is successful.
Figure 9: An example result of 3D processing followed by 3D plane
segmentation.
Figure 10: A second example result of 3D processing followed by 3D plane
segmentation.