Related Experiment Video
Updated: Jul 21, 2026
05:12
Robotized Testing of Camera Positions to Determine Ideal Configuration for Stereo 3D Visualization of Open-Heart Surgery
Published on: August 12, 2021
2.5K
Lightweight LiDAR-Based 3D Human Pose Estimation via 2D Depth Images for Autonomous Driving
Gyu-Yeon Kim1, Somi Park2, Sunkyung Lee2
1Graduate Program in Smart Factory, Ewha Womans University, Seoul 03760, Republic of Korea.
Sensors (Basel, Switzerland)
|March 14, 2026
Summary
This study introduces a lightweight 3D human pose estimation method using LiDAR for autonomous driving. It enhances safety by efficiently predicting pedestrian movements, even with self-occlusions.
Area of Science:
- Robotics and Computer Vision
- Autonomous Systems
- Sensor Fusion
Background:
- Pedestrian pose estimation is vital for autonomous driving safety.
- LiDAR sensors offer accurate 3D data and privacy benefits for pose estimation.
- Existing LiDAR methods are computationally intensive, limiting real-time application.
Purpose of the Study:
- To develop a computationally efficient LiDAR-based 3D human pose estimation method for autonomous driving.
- To address the high computational cost and memory requirements of current methods.
- To improve robustness in challenging poses like side-views and bending.
Main Methods:
- Projecting LiDAR point clouds into 2D depth images to reduce complexity.
- Utilizing a lightweight MoveNet model for initial pose estimation.
- Implementing an efficient 3D lifting technique.
- Introducing a self-occlusion correction algorithm for improved accuracy.
Main Results:
- Achieved competitive 3D human pose estimation accuracy.
- Significantly improved computational efficiency compared to direct 3D processing methods.
- Demonstrated robustness under side-view and bending poses through self-occlusion correction.
Conclusions:
- The proposed lightweight method is practical and scalable for real-time autonomous vehicle applications.
- The approach balances accuracy and efficiency, making it suitable for resource-constrained systems.
- Enhanced pedestrian detection and tracking capabilities in autonomous driving.
