Motion Recognition and Spinal Monitoring Based on Hidden Markov Models and K-means Clustering Using Wearable Sensors

As the internet of things (IoT) turns into the focus of software market, motion recognition has become popular in many areas, such as health care and real-time monitoring. Thus, this paper presents a motion recognition and spinal monitoring method using wearable sensors based on Hidden Markov Models (HMM) and K-means. To predict six different motions, i.e., walking, walking-upstairs, walking-downstairs, sitting, standing and lying down, we use inertial signals from waist-mounted smartphone sensors as inputs. And we put three sensors on the back to monitor the spine. When the motion is “sitting” and the spine curves to the right, left or forward more than 30 seconds then the system will give a warning beep. We use a forward-backward algorithm to train the hidden Markov models for each motion and use forward algorithm to choose the best fitting model among six different hidden Markov models. Furthermore, k-means clustering is used to cluster raw data into different classes. And at the same time, k-nearest neighbor (k-NN) algorithm is used to find the nearest classes that the raw data may belong to. The result of the motion recognition rate is 92.67% and the result of the spinal movement detection rate can be as high as 100%. Keywords—PERCLOS; Hidden Markov Model; K-means Clustering; K-Nearest Neighbor; Motion Recognition; Spinal Monitoring.