Depression Population Recognition Based On Intelligent Gait Analysis

With the advancement of urbanization,people's living standards have improved,and life pressures have become increasingly larger.This has led to not only the emergence of physiological problems,but also the emergence of psychological diseases.Depression,a mental illness that plagues 300 million people in the 21 st century,may not only bring daily physical and psychological torture to patients,but may also lead to pessimistic mood and suicidal behavior.Therefore,depressed people need to receive appropriate treatment as soon as possible to avoid the deterioration of the disease,and early screening and testing is particularly important at this time.Clinical assessment of traditional depression is made by a psychiatrist or clinical psychologist who diagnoses primarily based on medical history,clinical symptoms,duration of illness,and physical examination and laboratory tests.However,people with depression are less motivated and have difficulty getting medical care.In order to circumvent this difficulty and achieve better early screening for depressed people,this study explores the depression population recognition through intelligent gait analysis.The similar type of research is rare at home and abroad,so the research can be regarded as pioneering and cutting-edge at some point.The main research work is summarized below.(1)Designed a "no interference" gait information acquisition system based on multiple Kinect cameras,including hardware deployment design and software development and design.The deployment of the hardware uses multiple Kinect cameras mounted on the ceiling at a height of 2.65 meters from the ground,which greatly weakens the existence of the camera and plays the role of “no interference”.The acquisition software adopts the "master-slave" control and communication structure based on IP address,which satisfies the task of simultaneously controlling the simultaneous acquisition of multiple Kinect cameras by one button,and also has the functions of monitoring task status and data compression between devices.(2)Establishment of a multidimensional gait database for depressed and normal populations.The project research cooperated with the People's Hospital of **,and the specific program was deployed and implemented in the unit.The gait information of39 people with depression or those at risk of depression was collected.In addition,164 healthy graduate students were recruited and their gait information was collected as a control group.The acquired gait information data types include color videos,depth images,infrared images,and skeletons information.(3)Classification of depression and normal population based on gait information and comparative analysis of kinematic parameters.After pre-processing the collected human skeleton data by filtering and coordinate transformation,two traditional machine learning models,K Near Neighbor(KNN),Support Vector Machine(SVM),and two deep learning models,Long Short Time Memory(LSTM),Time Convolutional Network(TCN),are used to conducts end-to-end model training and prediction of gait data for normal and depressed populations.The classification correctness rates of the four models namely KNN,SVM,LSTM,and TCN are79.55%,89.39%,89.39%,and 85.61% respectively on test data set.In addition,the ROC(Receiver Operating Characteristic)curve,AUC(Area under Curve)value,accuracy rate-recall rate curve and AP(Average Precision)value are compared for the evaluation of models.LSTM outperformed other three models.In the comparative analysis of kinematic parameters,the depressed population had a smaller walking speed,a smaller average elbow joint angle,and a smaller average knee joint angle than the normal population,and the difference was significant.