Fine-grained Gesture Recognition By Using MEMS Sensors

Gesture recognition refers to the recognition of human behavior through a computer.With the continuous development of computer technology and the improvement of people's life quality,gesture recognition plays an increasingly important role in the Internet of Things(IOT).Gesture recognition can be widely used in smart home,smart medical and game interaction.The fine-grained gesture recognition can achieve better human-computer interaction,so that users can get a better experience.Gesture recognition can be divided into two categories,one is gesture recognition based on image or video.However,image-based gesture recognition technology is limited by light and line-of-sight,and requires a large amount of calculations and high deployment costs.The other is gesture recognition based on wireless signals.With the development of micro-electromechanical technology,miniature electromechanical inertial sensors with low cost,high precision and small size have appeared.Therefore,gesture recognition based on wearable MEMS inertial sensors has attracted wide attention.However,most of current gesture recognition methods based on MEMS sensors use machine learning methods to classify gestures by template matching.This will result in limited types of gesture recognition and higher latency of the algorithm.Even for the same gesture that has already been trained,different gesture amplitude may result in failure to recognize.In order to overcome these drawbacks,we propose an approach,which will be able to track the human body motion in real-time and also recognize complicated gestures.The system includes an inertial sensor unit and a wireless communication module for transmitting data to the host computer.Then,we correct the hardware and de-noise the motion data by smoothing and removing the gravity acceleration.Finally,the position of the inertial sensor is calculated,then the position and the human body model are used to predict the final gesture of the human body.After 40 round of different gesture tests,experimental results show that,the successfully recognition rate of our algorithm is 100%.The tracking accuracy of human body motion is only about 0.06 m.We also identified twenty gestures with different amplitudes.Compared with the traditional gesture recognition algorithm,our recognition success rate can be increased by 60%.So our approach can recognize the same gestures with different amplitudes.