Design And Applications Of A Wearble Human Daily Action Recogniton System

Human action recognition technology has a very broad application in our daily life,and plays a huge role in sports training,medical rehabilitation,film and television program production,and falling detection.Therefore,the field has always been a research hotspot.Currently,the most widely used method for human action recognition is computer vision.However,the vision-based action recognition system is easily affected by the environment disturbance,invades personal privacy,and the camera position must be fixed and the portability is poor.In order to overcome this shortcoming,people gradually consider alternative way for human action recognition.With rapid development of micro-electro-mechanical system(MEMS)technology,the size of the sensor is decreasing and the price is reducing.The development of wireless transmission technology makes it possible to construct micro-sensor networks for our daily life.All these provide a basis for the development of human action recognition system based on wearable devices.So in this paper we recognize human action using the wearable devices.This paper firstly designed a wearable sensor system for collecting human action information.The IMU is used to acquire action information such as acceleration,angular velocity,and posture.Considering that human motion contains rich biomechanical information,in addition to acceleration information,foot pressure,myoelectric information,heart rate information,etc.can also be used to describe the movement of the human body.Therefore,the motion information collected by a single type of sensor cannot fully describe the human body motion.Walking is the process of continuous interaction between the sole and the ground.Since the movement of the human body is basically accompanied by the change of the pressure of the sole,the movement of the human body is closely related to the sole pressure.Therefore,in addition to the IMU,this paper also selects a wearable foot pressure measuring device to collect the pressure information.Then,according to the human skeleton model and foot pressure distribution,the placement positions of IMU and pressure sensors are determined.Finally,the collected information is sent to the PC through wireless transmission device for action recognition.Then,we process the collected data.Firstly,the collected action sequences are smoothed by moving averaging to reduce the influence of noise.Considering that the collected action data contains a large amount of static action data,a threshold-based action segmentation algorithm is designed to extract useful action segments.After the useful action segment is extracted,the common action features are extracted from the statistical features and the physical features.Finally,we use decision tree algorithm and support vector machine(SVM)algorithm to recognize 13 pre-defined actions.Action features are often selected by experience,and redundancy often occurs in the process of feature selection,which will increase the computational complexity when too many features are selected.Firstly,decision tree algorithm is used to recognize actions.In the process of decision tree recognition,useful features will be screened out,thus reducing the number of features.Then the features selected by the decision tree are used in the SVM algorithm,and the experiment shows that SVM algorithm has achieved good recognition results.