PCViD project:
End-to-end system of object detection and recognition that takes as input sparse point clouds collected by cheap and easy-to-deploy sensors. RGB-based object detection models operate directly on RGB images, since they are dense in pixel space and encapsulate notions such as shape and texture.
This is not the case for sensor point clouds. Therefore, a crucial part for the success of sensor-based object detection is the development of reliable shape reconstruction/recovery methods from point clouds.