TY - JOUR AU - Zereen, Aniqua Nusrat AU - Gurung, Anubinda AU - Rajak, Amir AU - Moonrinta, Jednipat AU - Dailey, Matthew Nelson AU - Ekpanyapong, Mongkol AU - Vachalathiti, Roongtiwa AU - Bovonsunthonchai, Sunee PY - 2021/11/25 Y2 - 2024/03/29 TI - Automatic Elderly Fall and Unstable Movement Detection System Using Framewise and LSTM Based Video Analytics on an Embedded Device JF - NBTC Journal JA - NBTC. J. VL - 5 IS - 5 SE - Academic article DO - UR - https://so04.tci-thaijo.org/index.php/NBTC_Journal/article/view/253616 SP - 117-134 AB - <p>We introduce an edge processing device and cloud computation framework enabling activity profiling, unstable motion alerts, and fall alerts for elderly people living at home under their families’ care. The system analyzes video frames captured by fixed cameras, tracking each visible person and classifying their actions into one of nine ordinary activities, a fall, or unstable movement. Alert notifications are sent to caregivers if a fall or unstable movement is detected. The system comprises an embedded device (NVIDIA TX2) and cloud-based storage and analysis infrastructure. The main modules include video-based human detection, tracking, and recognition; fall detection; activity classification; and detection of painful, unstable, and confused motion likely to lead to falls. The system is designed for accuracy, usability, and cost. Individual module tests and a field test with a volunteer household indicate that the prototype system is ready for the next stage of commercial exploitation, with an accuracy of 91.6% for ordinary actions and falls as well as a recall of 97.02% for unstable motion.</p> ER -