Resource-constrained onboard inference of 3D object detection and localisation in point clouds targeting self-driving applications