A Computationally Efficient Bad Posture Detection Method to Alert Computer Users in Real Time

These days lot of people spend long hours at a computer, so they face lot of health issues because of wrong sitting postures. Good posture is crucial for human well-being. A vision-based system is proposed to minimize the poor ergonomic practices among computer users. It can be installed in the corresponding user computer, and it runs with low computation burden. The proposed system is based on image comparison with a reference image. The reference image is an image, that the user utilizing their webcam or computer camera to capture themselves in the correct seated posture. Initially it has been a converted into grayscale and applied a Gaussian filter before store in the memory. After making the reference image, the proposed system is stared to capture user’s image at a free defined frequency. The captured user’s image is filtered and converted to a binary image using a predefined threshold. After that, two parallel processes are applied to same resultant image to identify the moment of left and right side and bending of forward and backward. The decision of both processes has been estimated by comparing the calculated black- to-white pixel ratio with a predefined threshold. The notification has been generated via a pop-up message on corresponding computer screen. To test the effectiveness, the proposed system has been implemented on a laptop computer in a Python environment with the support of OpenCV library. The test has been conducted of four different bad postures such as over leaning to the right angle, left angle, forward and backward postures with 150 attempts. The test results shows that it has high potential to identify the bad postures of computer uses. Keywords: Poor ergonomic practices; Image comparison; Gaussian filter; Binary image; Black-to- white pixel ratio