Healthy human sitting posture estimation in RGB-D scenes using object context

Abstract

Unhealthy sitting posture leads to cervical spondylosis and other related cumulative trauma disorders (CTDs). Unfortunately, the research on the investigation of heathy sitting posture is rare. The current research is to estimate heathy sitting posture based on a computer workstation ergonomics perspective. A novel RGB-D scene healthy human sitting posture estimation framework was developed to estimate the sitting posture, in which a human posture is represented by 15 skeletal joints. A healthy human sitting posture configuration is defined from the view of ergonomics, a Naïve Bayes classifier was used to learn the health-constrained spatial and context relationships between objects and the human skeletal joints in the RGB-D scene. At the estimation stage, the object spatial features (e.g., coordinate, distance, height and angle) in the RGB-D scene were obtained through conducting the scene labeling. 15 human skeletal joints were extracted simultaneously from Kinect as primary inputs, and then algorithms were developed to generate and to classify the candidate healthy skeleton joints. Through skeleton refinement, the skeleton joints distribution of a healthy sitting posture was produced. The framework was tested on a dataset comprised of RGB-D scenes, which were collected from 3 subjects (3 types of sitting postures, each in 3 different offices). The experiment results indicate that the framework is feasible and reliable.