Using a Rotating 3D LiDAR on a Mobile Robot for Estimation of Person's Body Angle and Gender

Area of Science:

• Robotics
• Computer Vision
• Human-Robot Interaction

Background:

• Light Detection And Ranging (LiDAR) is typically used for robot navigation and tracking.
• Estimating human attributes from LiDAR data is crucial for advanced human-robot interaction.
• Developing efficient methods for training human attribute estimators is essential.

Purpose of the Study:

• To investigate the use of a 3D LiDAR sensor on a social robot for estimating human body orientation and gender.
• To explore alternative data acquisition methods like simulations and lab data for training.
• To evaluate the performance of convolutional neural network-based estimators on real-world data.

Main Methods:

• Utilized a rotating multi-layer 3D LiDAR sensor (Velodyne HDL-32E) mounted on a social robot.
• Trained convolutional neural network (CNN) models for estimating body orientation and gender.
• Compared performance using simulated data, lab-collected data, and real-world public space data.

Main Results:

• Achieved usable estimation of body angle (mean absolute error 33.5°) using 3D LiDAR.
• Simulated data proved effective for training body orientation estimators.
• Gender estimation accuracy exceeded 80% at close range, but required real-world data for training.

Conclusions:

• 3D LiDAR sensors are viable for estimating human body orientation in social robotics.
• Simulated data offers an efficient training approach for orientation estimation.
• Accurate gender estimation using LiDAR requires actual human measurements, especially for robust performance.