Sensor Simulation

MultiTest employs a physical-aware sensor simulation module to render images and point clouds for the object to be inserted. Here we provide more details about sensor parameter configuration and visualisation of occlusion handling.

Sensor Parameter Configuration

We leverage the configurations of Velodyne HDL-64E lidar and PointGray Flea2 color camera provided by KITTI to build our virtual simulator. Here we provide configuration parameters of our virtual sensor. Default means using the calibration of the sensors provided by KITTI.

Occlusion Handling

When inserting an object, occlusion should be carefully addressed. For instance, the inserted object should not block any objects that are closer to the sensor according to the distance to the sensor.

The details of occlusion handling operation are as follows

First, we obtain the minimum convex hull of the object by projecting it according to its position in the LiDAR coordinate system.
Next, based on the position of LIDAR and the convex hull of object, we could construct a vertebral-like region, which contains a visible region and a invisible region .
Finally, we remove all point clouds locate in the invisible region of the pyramid-like region.

Google Sites

Report abuse