Default SLAM

Post-Processing SLAM Data with PCMasterPro #

Getting Started #

In order to run SLAM projects, SLAM must be turned on in PCMasterPro. Open PCMasterPro and select the “SLAM Processing Tab.” Choose the appropriate profile settings: indoor, outdoor, or custom, as shown in Figure 2.

Notes:

• To revert to normal PPK processing, select “Disabled” in the SLAM settings before opening an existing project.
• A network connection is required only for validating the SLAM license key; it is not needed during cloud processing.

Available Processing Settings: #

• Min Matches to Relocalize & Min Matches to Track: Users should be able to change these settings based on the scan environment. If the scan is taken in an indoor region, features may be limited or sparse and thus a lower value is desired for this category in comparison to outdoor scenes. This relaxes the processing constraints and allows keyframes (each complete lidar revolution scan) to process easily with lower constraints on feature matching.
• Maximum Match distance frame to Map coefficient and Maximum match distance frame to frame: These settings adjust the scale factors for frame-to-frame translation and frame to map translation. The scan environment and application will require these parameters to be changed by the user. Note that as a rule of thumb the frame-to-frame scale factor is 4 times the frame to map scale factor for generating feasible results for all applications and scan regions.
  • Recommended: (Default tuning values (confirmed from testing) are noted below)
    • Small indoor room: frame to map SF – 0.1, frame to frame SF – 0.4
    • Large indoor room: frame to map SF – 0.25, frame to frame SF – 1.0
    • Small outdoor area: frame to map SF – 0.8, frame to frame SF – 3.2
    • Large outdoor area: frame to map SF – 1, frame to frame SF – 4
• Minimum Point distance & Maximum Point distance: This setting controls thresholds for allowing the algorithm to know which points need to be accepted/utilized for processing in terms of how far the points to be scanned are from the device. For e.g. If you are in tighter regions, lower the minimum point distance to capture ranges from features that are within close proximity.
• Tracking voxel size: This setting will change based on the number of points being collected (area of scan region) and how accurate the user would want the final trajectory to be at the cost of compute time and power. Since the algorithm will be storing dense keyframes in the backend, the preferred voxel size becomes an important tradeoff to consider between processing time and accuracy. Using a smaller voxel size will increase the number of tracking points in each processing key frame but significantly increase processing time. Likewise, using a high voxel tracking size will reduce accuracy of the final point cloud (may introduce greater relative inaccuracy between features) but decrease processing time. It is recommended to use 0.1 m for indoor applications and 0.2-0.3m for outdoor applications.
  • NOTE: A smaller voxel size will generally produce a more precise point cloud but at the cost of processing time.
• INS Tracking Mode: If the scanned motion path consists of jerky movements, rapid turns, and varying accelerations; one can enable the INS Tracking Mode checkbox before processing the cloud for enhancing the SLAM’s tracking performance to output a more accurate trajectory that considers non-linear motion more carefully. 
• Undistort with LiDAR + IMU:  If a static alignment during the IMU initialization phase was fairly stable then the user can enable “Undistort with IMU” and “Undistort with LiDAR Motion”. At times if the static alignment of the IMU may have been unsuccessful, the user can reprocess the cloud by just selecting “Undistort with LiDAR motion” only (and unchecking the stationary IMU alignment setting) which will ensure that the IMU readings aren’t used for motion blur correction.
• Use Surfels: Enabling this setting will allow the SLAM algorithm to examine planar surfaces (e.g. windows) in addition to continue tracking features using a voxelized map of the environment. This setting enhances the tracking ability of features within the environment and is useful to apply when processing indoor or urban region scans.
• “Weight of IMU Prior” – This value shouldn’t exceed 0.05 (0.02 by default), it is the weightage given to IMU pose estimates during regions in the scan when the LiDAR tracking state is determined to be less reliable by the internal SLAM sensor fusion algorithm.
• “Max Frames Track Using Only Predicted Pose” – An integer value is accepted for determining the maximum number of frames the SLAM algorithm should continue tracking using only a predicted position estimated by the IMU. This is useful for edge cases when the LiDAR tracking state is lost, e.g. a long corridor with no observable features.
• Maximum Laser Return ID to use: This field identifies the specific laser return type to use for SLAM processing from the LiDAR.

• • For Hesai XT-32: This value should be set to 2 (default)
  • For Hesai XT32-M2X: This value should be set to 3 (recommended)
• Georeference Option – To georeference the collected SLAM point cloud with calculated GNSS positions, switch this setting ON. 
  • The SLAM point cloud can only be georeferenced if the mission was collected with a GNSS antenna connected to the RESEPI unit during the mission. More information can be found here  

Additional Application Notes:

• If the unit did not go through an adequate amount of IMU initialization time (i.e. Unit was not completely static or the time frame for the initialization time was under 8-10 seconds) then it is worth disabling the ‘Stationary IMU alignment’ setting and using only Undistort with LiDAR Motion.
• Other IMU Specific settings should ideally be utilized only when the default settings aren’t sufficient to build a high quality, dense point cloud.