Minimum LiDAR Point Density
Minimum LiDAR Point Density is a sanity check that automatically monitors the number of LiDAR points captured within each cuboid annotation relative to its distance from the ego vehicle. When the point count falls below the project-configured threshold for that distance range, the system surfaces a non-blocking warning - enabling labellers to make informed decisions about annotation confidence without interrupting their workflow.
Point density thresholds are configured at the project workflow level. If not defined by the customer, iMerit can recommend suitable values based on the sensor suite and operational experience.
Use Case
In LiDAR-based annotation, distant objects return progressively fewer points, making accurate cuboid placement inherently harder. Without a clear guideline on what constitutes a valid annotation at a given range, labellers have no objective reference to flag ambiguous or under-represented objects.
Minimum LiDAR Point Density addresses this by:
Tracking the real-time point count inside every active cuboid
Comparing it against the distance-dependent threshold defined in the project workflow
Issuing a per-track warning referenced by track ID and frame number - whenever the guideline is not met
Common scenarios include:
Distant vehicles or pedestrians at the boundary of sensor range
Partially occluded objects with sparse returns
Night or adverse-weather captures where density is reduced across the board
Quality audits where reviewers need a systematic record of low-confidence frames
Benefits
For Annotators
Immediate, Non-Blocking Warnings - Alerts appear only when guidelines are violated, so the workflow is never interrupted unnecessarily.
Frame-Level Precision - Each warning cites the specific track ID and frame number, letting annotators jump directly to the problematic annotation.
Confidence Calibration - Provides an objective signal to distinguish genuinely sparse objects from misplaced cuboids.
Reduced Guesswork - No need to manually count points or cross-reference distance tables; the tool does it automatically.
For Project Managers
Customer-Aligned Quality Standards - Thresholds can mirror customer guidelines exactly, ensuring deliverables meet contractual expectations.
Sensor-Aware Configuration - iMerit can pre-configure recommended thresholds per sensor suite, reducing onboarding time for new projects.
Audit Trail - Warnings are surfaced per track ID and frame, making QA reviews faster and more systematic.
Reduced Rework - Problems caught at labelling time cost significantly less to fix than those found during review.
Steps to Use
Configure Thresholds in the Project
Navigate to your project's recipe settings (category schema).
Locate and enable the Sanity Checks section.
Define distance bands (e.g., upto 30 m, 30 - 60 m, 60 - 100 m, 100 m+).
Set the minimum point count threshold for each band per object class.
Save and publish the recipe. Thresholds apply immediately to all active tasks on reload.
Respond to the Visual Cue
If a cuboid's point count falls below the configured threshold, a visual cue appears on all the affected cuboids on the timeline.
The cue identifies the frames where the violation was detected.
Determine whether the object is genuinely sparse (far / occluded) or whether the cuboid is misplaced.
The cue clears automatically once the point density is within the permitted range.
Best Practices
Start with Conservative Thresholds - Begin with higher minimum point counts and relax them based on annotator feedback and QA outcomes.
Band by Sensor Range, Not Arbitrary Distances - Align distance bands with the known effective range of the sensor suite (e.g., dense near-field, sparse far-field).
Per-Class Customisation — Pedestrians and cyclists present fewer points than large vehicles at equivalent distances; set thresholds accordingly.
Document Exceptions - For sequences with known sensor degradation (fog, rain, direct sunlight), consider creating a new project for such degradation and adjust thresholds rather than globally on the same project.
Coordinate with iMerit - If you are unsure of appropriate starting values, request an iMerit recommendation based on your sensor specifications.
Review Warning Patterns - A high volume of warnings on specific distance bands may indicate that either the threshold or the annotation approach needs calibration.
Last updated