3D LIDAR Birdview Annotation for Lower-Cost ADAS Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vehicle advanced driver assistance systems (ADAS) that utilize LIDAR for object detection face challenges in distinguishing between different types of objects without extensive deep neural network training, requiring substantial manually annotated data, which increases costs.
Innovation Solution
A semi-automatic 3D LIDAR point cloud annotation system that filters, normalizes, and quantizes LIDAR data into 2D birdview projections, allowing human annotators to identify objects, with the system then converting these annotations back into 3D data for improved model training, potentially reducing the need for further human input.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If extensively trained deep neural networks are used to distinguish different types of objects, then object detection accuracy is improved, but the amount of manually annotated training data required increases substantially
Solution Approach 1:
The system performs preliminary filtering of LIDAR data to remove non-object points (road surface, sky, vegetation) before annotation. This preprocessing step reduces the volume of data requiring manual annotation while preserving all object-related points, thereby maintaining detection accuracy with less annotated data
Solution Approach 2:
The system generates synthetic annotated data by projecting 3D LIDAR points to 2D images, applying image annotation, and back-projecting to create synthetic 3D annotations. These synthetic annotations serve as copies that can be used for training without requiring equivalent manual 3D annotation effort
2Measurement precision
If manually annotated 3D LIDAR point clouds are obtained for training, then object detection capability is improved, but annotation costs increase
Solution Approach 1:
The system creates synthetic annotated 3D LIDAR data by copying and transforming 2D image annotations into 3D space. This allows generation of training data with minimal human effort, dramatically reducing annotation costs while providing sufficient training examples for object detection
Solution Approach 2:
The system automates the annotation process by using the LIDAR point cloud structure itself to generate candidate object regions and infer 3D bounding boxes from 2D annotations. This self-service capability reduces reliance on expensive manual 3D annotation while maintaining data quality
3Measurement precision
If full manual annotation of 3D LIDAR data is performed, then annotation accuracy is improved, but annotation time increases
Solution Approach 1:
The system performs preliminary clustering of LIDAR points into candidate object regions before human annotation. This pre-organization of data presents annotators with pre-grouped points that are likely to be objects, reducing the time required to identify and annotate objects while maintaining accuracy through focused human review
Solution Approach 2:
The system generates synthetic annotations that can be copied and applied to similar scenarios. Once annotators label one instance of an object type, the annotation can be copied or adapted to similar objects, significantly reducing repeated annotation time while maintaining consistency and accuracy
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach reduces the reliance on extensive manual annotation, lowering costs and enhancing the efficiency of object detection in ADAS systems by utilizing semi-automatic annotation methods.
Implementation Method 1
a LIDAR system configured to emit laser light pulses and capture reflected laser light pulses from a vehicle collectively forming 3D LIDAR point cloud data surrounding the vehicle
Implementation Method 2
LIDAR systems emit laser light pulses and capture pulses that are reflected back by surrounding objects
Data Source
AI summary
A semi-automatic three-dimensional light detection and ranging (LIDAR) point cloud data annotation system and method for autonomous driving of a vehicle involve filtering 3D LIDAR point cloud and normalizing the filtered 3D LIDAR point cloud data relative to the vehicle to obtain normalized 3D LIDAR point cloud data, quantizing the normalized 3D LIDAR point cloud data by dividing it into a set of 3D voxels, projecting the set of 3D voxels to a 2D birdview, identifying a possible object by applying clustering to the 2D birdview projection, obtaining an annotated 2D birdview projection including annotations by a human annotator via the annotation system regarding whether the bounding box corresponds to a confirmed object and a type of the confirmed object, and converting the annotated 2D birdview projection to back into annotated 3D LIDAR point cloud data.