matlab-import-driving-data

Load raw GPS, camera, lidar, and annotation files, then construct scenariobuilder.GPSData, CameraData, LidarData, and ActorTrackData objects with matching timestamps. Always run convertTimestamps and normalizeTimestamps after GPSData construction so downstream synchronize and export APIs work correctly.

Convert all sensor timestamps to numeric seconds, pick the longest-duration sensor as reference with normalizeTimestamps, align the others to that reference, then call synchronize between objects. Optionally crop all sensors to a common time window.

Supported inputs include KITTI, nuScenes, Waymo, Pandaset, ROS and ROS2 bags, MAT files, CSV, and MP4 video. Point clouds can be read from PCD, PLY, BIN, NPY, and LAS/LAZ formats, and GPS from JSON, HDF5, CSV, or ROS bag topics.

Yes, actor tracks can be computed from lidar using deep learning detectors like PointPillars or a classical clustering-plus-tracking pipeline. Results depend on scene complexity, and deep learning detectors perform much better than clustering in dense urban scenes.

drivingLogAnalyzer is opt-in only, launched when you explicitly request it or report sensor problems like sync drift or timestamp mismatch. For building scenarios or exporting to RoadRunner, OpenSCENARIO, or Unreal, hand off to the matlab-use-scenario-builder skill instead.

DLA requires SensorParameters to be a struct with four fields: MountingLocation, MountingAngles, Intrinsics, and a positive EgoOriginHeight. A bare cameraIntrinsics object is accepted by the property setter but the overlay refuses it at runtime.