Research On Inertial Sensor Based Sensing Scheme For Human Activity

Nowadays,inertial sensors have become the standard accessories of smartphones and wearable devices.One inertial sensor mainly consists of an accelerometer and a gyroscope,which can measure acceleration and angular velocity.By calculating we can further get more data,such as velocity,displacement,turning angle,etc.Inertial sensors have enriched functions of smartphones and wearable devices.Inertial sen-sors have been widely used in localization and navigation,fitness monitoring,human-computer interaction,kinematic analysis,etc.Consequently,inertial sensor based hu-man activity sensing has become a hot area of research in recent years.In this paper,we research on inertial sensor based human activity sensing from the following two aspects.The first is inertial sensor based objective activity recognition.Objective activity means the activity which the user wants to sense.For example,the objective activity is walking and turning in inertial sensor based localization and navigation,while the objective activity in inertial sensor based fitness monitoring is the appointed activity of the user.We first research on step recognition.We find that the change of inertial data of pedestrian walking are periodic,so we can recognize step by detecting peri-od.Then we research on turn recognition.We detect turning by angular velocity and get turning angle by integration.After that,we research on motion recognition.We can classify motions by template matching.On this basis,we design a human-drone interaction system based on the recognition of objective activity.In this system,we employ wearable devices to sense the objective activity to realize interaction.Besides,we implement our system with commercial products.The second is inertial sensor based abnormal activity detection.In the recognition of objective motion,there may exist abnormal motion which is really different from objective motion.However,current objective motion recognition mechanism can't judge whether a testing motion belongs to the range of objective motion or not.That is to say,the abnormal motion will also be classified into some class of the objec-tive motions.It's inapposite in applications such as human-computer interaction and kinematic analysis.In this paper,we research on how to detect abnormal human activ-ity based on inertial sensor and propose a new method for abnormal motion detection based on semi-supervised learning.Our method extracts meaningful motion features from inertial sensor data and can effectively differentiate between abnormal motion-s and objective motions.Experiment result shows that our method can achieve 89%accuracy.