Design of a Computer Vision System for Human Pose Tracking Within the Aircraft Cabin

Passenger comfort is a deterministic factor in today’s competitive air travel industry and particularly for business jets which are increasingly being designed and adapted to maximize comfort for their occupants. Sitting posture could be used as an effective way of gaining insight into passenger comfort. Despite which very few studies have focused on tracking sitting posture in aircraft passengers. Given recent advancements in Computer Vision (CV) and Deep Learning (DL), open-source pre-trained 3D Human Pose Estimation (HPE) algorithms are readily available and can be used to estimate human pose in various conditions. Thus, the primary aim of the thesis was to explore the application of the latest vision algorithms for passenger pose estimation in aircraft cabins and design a prototype for real-time seating pose detection. This entailed use of a pre-trained, open-source, 3D Human Pose Estimation algorithm, MediaPipe Pose, which detects humans in an RGB image, localizes their joint locations and outputs 3D pose landmarks (keypoint locations) in real-time using MediaPipe framework and OpenCV library. This prototype was tested in a real aircraft cabin and the results from the pose estimation were used to draw 3D animation to confirm real-time pose tracking. The system was able to estimate body poses regardless of sitting posture (active, passive, lateral bending to left/right). As such, this thesis proposes a prototype for real-time vision-based 3D pose tracking in aircraft passengers.