Detection and Low-Latency Notification of Improper Backpack Posture using Deep Learning

Abstract

Research shows that school children often carry overweight backpacks which make them perform improper posture, leading to thousands of musculoskeletal injuries each year. Key elements of carrying packs properly include that the pack’s weight should be no more than 15% of a child’s body weight, the pack’s weight should be distributed evenly, and children should walk upright without leaning side to side or bending front or back due to any form of weight compensation. This study focuses on combining force-sensitive resistors, accelerometers, and load cells to detect children’s posture and give feedback on whether the pack is overweight or whether the child walks properly when carrying the pack. This study uses supervised machine learning for posture detection and successfully obtains a high accuracy of 90.71% in posture classification. The prototype aims to be a building block to a more accessible and affordable package in developing nations.