Recognition of Sitting Posture Using Deep Learning and OpenCV

Because of its essential issues and wide range of applications, vision-based human posture identification has piqued the interest of many researchers. The applications in surveillance systems range from simple posture recognition to complex behaviour comprehension. This leads to significant advancements in techniques for representing and recognising human body position. The applications and general framework of human body position identification are discussed in this research. We provide a method for detecting body sitting posture in OpenCV using OpenPose CNN technology in this study. The report also highlights its benefits and drawbacks. The domain of human body posture recognition has been active for over two decades and has produced a substantial amount of literature. The report also discusses some different approaches for detecting human position, as well as their pros and shortcomings. Adolescents' health can be harmed by sedentary behaviour and bad seating position. As a result, it is highly practical to efficiently identify and warn pupils about their sitting posture in the classroom. This research suggested an OpenPose-based in-class student sitting posture identification system that leverages the classroom monitor to identify students' sitting position and OpenPose to extract the posture feature. The convolutional neural network, which is used to train datasets and distinguish sitting posture of Body, is built using the Keras deep learning framework.

Human health is harmed by poor sitting posture. People can be reminded to correct their sitting posture using an intelligent sitting posture recognition algorithm. A seated pressure image acquisition system was devised in this paper. We introduced a unique hip positioning algorithm based on hip templates with the system. The algorithm's average variance for hip placement is 1.306 pixels (corresponding distance: 1.50 cm), and 94.1 percent of the maximum positioning deviation is less than three pixels. Statistics show that the algorithm performs admirably for a variety of subjects. At the same time, the algorithm is capable of not only accurately locating the hip location with a small rotation angle (0–15), but also adapting to the sitting posture with a medium rotation angle (15–30). Alternatively, a large rotation angle (30–45) might be used. The regional pressure values of the left hip, right hip, and caudal vertebrae are efficiently retrieved as features using the hip positioning algorithm.

Keywords: categorization of sitting posture, CNN, OpenCV, Open Pose