1. Insights
  2. AI Data
  3. Case Study

Building high-quality 3D flash lidar datasets for autonomous vehicles

Learn how TELUS International helped a leading automotive lidar systems provider create high-quality flash-lidar 3D datasets with our fully-managed data annotation solutions.

Cars with bounding boxes to show image annotation
  • Share on X
  • Share on Facebook
  • Share via email

99.55% recall and precision rate

The challenge

The client required 3D annotation tools compatible with their proprietary flash lidar technology, along with robust project management and a secure workforce. This included object detection support using 3D cuboid tools and lane marking annotations to determine drivable areas using 3D point cloud segmentation.

Solid-state flash lidar produces more dense data than traditional automotive lidar setups that generate image frames via raster scanning (i.e. one laser pulse per pixel at a time). Flash lidar, however, generates an image using a single laser pulse. These devices use a laser-based 3D imaging system where each laser pulse illuminates a large area, and a focal plane array (FPA) simultaneously detects light from thousands of adjacent directions. The 3D tools for annotating these dense datasets require a high level of sophistication to ensure accuracy, along with a configurable platform to seamlessly store and process such data.

The TELUS International solution

  • High-precision 3D labeling tools: Our robust 3D cuboid and segmentation tools supported class detection for objects like moving vehicles, pedestrians and drivable areas by focusing on retro refractor points (Retro refractor points are calibrated for shiny surfaces like metallic plates/objects).
  • Labeling automation features: 3D labeling features like one-click cuboids, aggregated point cloud mode, default dimensions and optimized tool shortcuts helped the client achieve high rates of annotator productivity.
  • Annotator training/qualification: Our secure and highly effective training and qualification program helped annotators execute unique and complex 3D annotation requirements while maintaining high accuracy levels.
  • Robust quality assurance: Our platform’s built-in heuristic check options, flexible scoring rubrics, multi-user verification workflow and quality analytics facilitated accurate and high-quality annotations
  • Custom feature engineering: Our team provided support to build custom features that provide a better visual display of the point intensity for the lane markings and retroreflectors to simplify 3D labeling.
  • Scalable solutions: Our flexibility to change the scope of guidelines, schema and workforce options based on the client’s ad-hoc requirements supported quick iterations without disrupting the client’s data pipeline.
  • End-to-end project management: Our experienced project managers handled all aspects of the project including data preprocessing, workforce planning/management, quality assurance and on-time deliveries.
  • Diverse workforce sourcing: Our global sourcing team leveraged our vast network of community members, helping to combat potential model bias through a diverse workforce.

The results

TELUS International delivered high-quality annotations at a 99.55% recall and precision rate for 3D object detection and segmentation projects. The client used these datasets to improve automotive perception via flash lidar technology.


Check out our solutions

Test and improve your machine learning models via our global AI Community of 1 million+ annotators and linguists.

Learn more