OpenPose, which works on edge devices, as IP cameras and AI Box set up near production lines, detect worker’s skeleton and pose in a factory. Through skeleton data OpenPose captures, TensorFlow judges if a pose is normal or abnormal at the primary level.
Data captured by edge devices is sent to MQTT broker in an edge server, then MQTT data is registered in Elasticsearch. Unusual skeleton data detected through unsupervised machine learning is sent to VANTIQ, using Webhook.
All of datum gathered to VANTIQ Edge is checked again if it’s normal or not. Datum are recorded along with captured images in VANTIQ Cloud and used for anomaly detection’s administration and notification. A dashboard is provided as well, on which recorded datum can be seen.