create_dataset

Create your own dataset

create_dataset/ contains scripts that show our pre-ML radar and lidar processing on raw sensor data. Use this only for creating your own radar-lidar images dataset (similar to dataset_5) to train with our models. You can ignore files in this folder if you do not want to create your own dataset.

• First, move your raw radar and lidar data to this folder in a similar folder structure as our raw dataset.

• timestamp_check_radar_lidar.py checks if the radar and lidar timestamps make sense.

  • We assume that each trajectory was recorded with the lidar capture starting first, radar starting next, and radar finishing first followed by lidar.
  • The output of this is a global common start and end timestamp for both the sensors. This gets stored in timestamp_files_radar_lidar/.
  • Depending on the original timestamps of each sensor, one may need to account for timestamp format, time zone offsets etc.

• create_dataset_all_radar_lidar.py uses the timestamped raw lidar point cloud and raw radar frames, and the global time start and end computed by timestamp_check_radar_lidar.py.

  • Lidar frames are captured at 20 Hz. We use the lidar timestamps as is and find the closest radar timestamp and frame to each lidar frame.
  • We then associate these two sensor frames and perform processing on each of them.
  • We convert the lidar point cloud to a high resolution greyscale image that can be a label for training.
  • We perform range-doppler-azimuth processing to the raw radar frame. We then perform a magnitude based thresholding. We then convert to a polar grey-scale image that can be used as input for the network. Our paper discusses this pre-ML processing in detail in Section. 3A.

• We assume that the radar and lidar packets from the sensors have been processed to be in a structure that is similar to our raw dataset.