An Improved Motion Intent Recognition Method For Intelligent Lower Limb Prosthesis Driven By Inertial Motion Capture Data

Prostheses which help most amputees address the movement obstacles play a vital role in the amputee's life.Intelligent lower limb prostheses can replace the losing parts of transtibial amputees to some extent,and help them recover self-care and work ability.Human motion intent recognition is particularly important in the research of intelligent lower limb prosthesis.Accurately and timely identifying the motion intent can provide automatic,natural,and seamless movement for transfemoral amputees.At present,there are many researches focus on the motion intent recognition of intelligent lower limb prostheses.Existing researches usually select appropriate features based on prior knowledge from bio-signals which collected by sensors embedded in prostheses,and implement intent recognition through pattern recognition.However,most traditional methods involve multiple types of sensors,which will introduce additional problems such as multi-source data fusion and heterogeneous data interference.Secondly,these studies often classify the patient's motion intention based on the motion capture data generated by the affected side after the transition,which may cause the lags of transitional state recognition.Therefore,we mainly focus on above shortages and propose two classes of improved motion intent recognition method based on intelligent lower limb prosthesis driven by inertial motion capture data.In Chapter 2,this dissertation firstly redefines the movement mode of intelligent lower limb prosthesis,especially the motion conversion mode,to solve the hysteresis of intent recognition.Then,before the transition of the affected side,only the time series data generated by the inertial sensor bound to the healthy side in the early stage of the swing phase is used as data sample.The statistics such as mean and variance and support vector machine are used for feature extraction and classification.The experimental results show that the method can accurately predict 13 classes of motion intent of lower limb amputations.Besides,both the traditional methods and the method proposed in Chapter 2 use manual features,which depend on expert experience.Due to the superiority of deep neural network which can automatically learn the low-level to high-level features,this dissertation uses self-learning features instead of manual features in Chapter 3,and proposes a new method for intent recognition.Specially,a convolutional neural network is used to establish the mapping between the state of motion of the healthy leg and the intent of the disabled in a data-driven manner.The experimental results show that the method can automatically learn the features of intent recognition data and accurately predict the motion intent based on intelligent lower limb prostheses.