Human posture recognition and good posture recommendation

Abstract

This study presents a model human posture recognition method using skeleton tracking. This is achieved using the power of Kinect, a motion sensing input device by Microsoft for the Xbox 360 video game console and a combination of posture recognition algorithm. The project focuses on development of a prototype that analyses posture of a human object in a scene, recognizes the posture (standing or sitting) and provide recommendation on a good posture. This has been achieved using a 2-stage head detection process to locate the people in a particular scene. This first explores the boundary information embedded in the depth map of Kinect to locate the candidate regions that may indicate the appearance of people. The algorithm implemented here scans across the whole image and gives the possible regions that may contain people. The regions are examined each using a 3D head model, which utilizes the relational depth information of the array for verification. Then parameters of the head are extracted from the depth array and use them to build a 3D head model. Later, a region growing algorithm is also developed to find the entire body of the person and extract his/her whole body contour. This project has been able to highlight how computing can be used to solve problems related to human behavior analysis. Particularly in scenarios where human behavior needs to be monitored in places such as Correctional facilities, Interogations, Classrooms, Healthcare, surveillance among others. On good posture recommendation, this method points out that, it is possible to utilize the power of computer vision in promoting correct human posture by providing a tool that tracks human posture, asses the time consumed in that posture and alerts the person in order to reduce muscle strain.