Study Of Human Behavior Recognition Based On Multi-feature Fusion

With the continuous development of the MEMS technology,human behavior recognition technology based on sensors has been developed rapidly.Human behavior recognition technology has been widely valued by all sectors of the community.Human behavior recognition has been widely used in many fields,such as motion detection,rehabilitation,assessment of energy consumption and so on.Compared with the human behavior recognition based on computer vision,the sensor based method can reflect the nature of the movement better,and has the advantages of portability,comfort,cost and easy maintenance.Although human behavior recognition technology based on sensors has made considerable progress,but there are still some key problems need to be solved,such as how to extract the appropriate characterization,how to construct a high accuracy and low complexity classifier etc.Focusing on these problems,the following work is done in this paper:1.The research status of human behavior recognition and gait phase identification is studied.The methods used in various research institutions is summarized,the advantages and disadvantages of each method is analyzed.At last,the method is determined according to the research target.2.A set of data acquisition device have been designed to acquire experimental data.The set is combined of two parts: accelerometer and gyroscope data acquisition device and Pedar-X distributed plantar pressure insole.The accelerometer and gyroscope data acquisition device is composed of STM32 microcontroller,MPU6050 gyroscopes and accelerometers and nRF24L01 wireless communication module.It can acquire acceleration and angular velocity real-time,and uploaded to the host computer.Pedar-X distributed plantar pressure pad can also acquire the pressure data in real-time and upload to PC.3.In the aspect of gait phase recognition,a new method based on multi-feature fusion for complex road condition is proposed.The gait phase identification of complicated road is regarded as the research object.The characteristics of the integration of the average plantar pressure,the center point of plantar pressure and the thigh angle were extracted to recognize different phases as initial stance,mid-stance,terminal stance,initial swing and terminal swing by DAG-SVM.4.In the aspect of human behavior recognition,a method of human behavior recognition based on multi-feature fusion is proposed in order to represent the behavior of human body better.The characteristics of acceleration and plantar pressure is combined in this method,and six kinds of behaviors is identified,such as standing,sitting,walking,going upstairs,downstairs and running by using hierarchical support vector machine.The overall recognition accuracy is above 92%.