Research On Motion Intent Recognition Of Intelligent Lower Limb Prosthesis Driven By Inertial Sensor Data

Motion intent recognition of lower limb prosthesis is an important prerequisite for correct and timely control of lower limb prosthesis motion,the goal is to provide accurate motion intent guidance for prosthesis wearers and help users to walk steadily and smoothly.According to the information sources,the methods of motion intent recognition of intelligent lower limb prosthesis are mainly based on two aspects: motion intent recognition based on bioelectrical signals and motion intent recognition based on biomechanical signals.Among them,the intent recognition of bioelectrical signals has the potential of human intuitive control,but it has not been widely used because of the difficulties in signal acquisition,the vulnerability to interference and instability.With the rapid development of sensor technology,inertial sensors can integrate various types of sensors(such as accelerometers,gyroscopes and magnetometers,and others),and intent recognition researchers widely use biomechanical signals to identify the motional intent of lower limb prosthesis.Traditional lower limb prosthesis intent recognition usually places sensors on the prosthesis.When the motion mode changes,there will be a lag in recognition,which will delay the control of the prosthesis.Prosthetic wearers will face the risk of falling.Therefore,this paper proposes a technical scheme of binding inertial sensors to the healthy side of a unilateral amputee.After redefining the motion pattern of the lower limb,the motion intent of the prosthesis can be identified by recognizing the motion pattern of the healthy side swing phase,and there will be no lag recognition in the conversion mode.Whether in the steady-state mode or in the conversion mode,lower limb motion is a continuous process,and lower limb motion intent recognition is a short-term action behavior in a moment.This paper mainly focuses on these two points and makes a preliminary attempt on data feature self-learning,and achieves certain recognition effect.Firstly,in the process of human motion,whether in time or space,the motion of lower limbs is continuous,but the data collected from inertial sensors are discrete.In order to better analyze the continuity of human lower limb motion,this paper proposes an intent recognition method based on time series data modeling in the second chapter.Firstly,using the method of Gauss mixture function,the motion intent data collected by inertial sensors are continuous.Then the continuous data is used to provide good data for parameter learning of the hidden Markov model of each motion intent.Finally,expectation maximization algorithm is used to effectively identify the motion intent of lower limb prosthesis.Secondly,due to the particularity of intent recognition in lower limb prosthesis,it can reflect the potential short-term human activity,and it's duration can't be more than one gait cycle.When the sample length is too short,the time-frequency domain features extracted by the traditional activity recognition algorithm are no longer stable,and then can't provide reliable intention recognition.Therefore,in the third chapter of this paper,an intent recognition method based on short-term data modeling is proposed.Firstly,moving intent data is segmented by sliding window,and an over-complete template library containing short-term activity of different human bodies with different intents is created.Then,1-norm is used to measure the similarity between the test template and the template library.Finally,the relaxation discriminant method is used to classify and recognize the intent of short-term motion.