Body and Head Orientation Estimation with Privacy Preserving LiDAR Sensors

Body and head orientation estimation is important in many applications such as pedestrian protection, movement prediction, robotics, and behavioral analysis. In this paper, we propose a system that uses privacy preserving LiDAR sensors to estimate body and head orientations of people, with a motivation of providing guidance feedback to individuals who face nonverbal social communication challenges in workplace settings, such as some individuals with Autism Spectrum Disorder (ASD). For example, people who tend to look away from a speaker could be coached on the importance of periodically making eye contact and showing overt attention, or could be discreetly provided with real-time feedback based on present behavior. We developed models for body and head orientation estimation, using low-resolution point cloud data from two LiDAR sensors. The body orientation estimation model uses an ellipse fitting method while the head orientation estimation model is a pipeline of geometric feature extraction and neural network-based regression. Compared with other body and head orientation estimation systems using RGB cameras, our proposed system uses LiDAR sensors to preserve user privacy, while achieving comparable accuracy. To the best of our knowledge, this is the first body and head orientation estimation system using depth sensors for which the sensors do not require a specified placement in front of the subject. Our model achieves a mean absolute estimation error of 8.4 degrees for body orientation and 16.5 degrees for head orientation.