Quanjun Song

HMM-Based Human Fall Detection and Prediction Method Using Tri-Axial Accelerometer

Falls in the elderly have always been a serious medical and social problem. To detect and predict falls, a hidden Markov model (HMM)-based method using tri-axial accelerations of human body is proposed. A wearable motion detection device using tri-axial accelerometer is designed and realized, which can detect and predict falls based on tri-axial acceleration of human upper trunk. The acceleration time series (ATS) extracted from human motion processes are used to describe human motion features, and the ATS extracted from human fall courses but before the collision are used to train HMM so as to build a random process mathematical model. Thus, the outputs of HMM, which express the marching degrees of input ATS and HMM, can be used to evaluate the risks to fall. The experiment results show that fall events can be predicted 200-400 ms ahead the occurrence of collisions, and distinguished from other daily life activities with an accuracy of 100%.

A research on automatic human fall detection method based on wearable inertial force information acquisition system

Frequent and high-risk fall accidents of the elders have become a serious medical and social problem. In this paper, a wireless automatic human fall detection device based on tri-axis accelerator is designed and realized; and a novel method to distinguish fall events from other daily activities is proposed also, including multi-impact falls and rolling down falls. In the real application, the device is worn on upper trunk of human body, with the algorithm focusing on the impact during a fall and the orientation of trunk before the fall and after it. At last, experiment is performed on some typical daily activities, such as walking, stand-to-sit, stand-to-squat, fall frontward, sideways, etc, and it has shown good results in high detection rate and low positive false rate.

A human identification method based on dynamic plantar pressure distribution

A novel human identification method based on dynamic plantar pressure distribution was proposed in this paper. An in-shoe plantar pressure measure system was applied to collect pressure information and the plantar pressure database was established. The wavelet transform was used to remove the noise. We extracted two sides of information including pressure amplitude and position during walking. The support vector machine (SVM) was employed to classify human. Experiments with samples of 35 persons showed that the recognition rate reached 96%, and the proposed method was insensitive for weight. However, the walking speed had significant influence on recognition performance.

Micromanipulator with integrated force sensor based on compliant parallel mechanism

Movement in micro-scale precision cannot be achieved by unaided human hand. This paper describes the design of a six degrees of freedom (DOF) micromanipulator based on compliant parallel mechanism. It is capable of delivering 6-DOF pure motions with high precision and featured by piezo-driven actuators, flexure hinges and integrated force sensor that can provide the system with real-time force information for feedback control. The static features of such a mechanism include high positioning accuracy, structural compactness and smooth and continuous displacements.

Research on the Surface EMG Signal for Human Body Motion Recognizing Based on Arm Wrestling Robot

In this paper, the surface electromyographic (EMG) signals is acquired from the upper limb when the experimenter competes with the arm wrestling robot (AWR) which is integrated with mechanical arm, elbow/wrist force sensors, servo motor, encoder, 3D MEMS accelerometer, and USB camera. The arm wrestling robot (AWR) is intended to play arm wrestling game with human on a table with pegs for entertainment and human upper limbs muscle modeling. As the EMG signal is a measurement of the anatomical and physiological characteristic of the given muscle, the macroscopical movement patterns of the human body can be classified and recognized. By using the method of wavelet packet transformation (WPT), the high-frequency noises can be eliminated effectively and the characteristics of EMG signals can be extracted. Auto-regressive (AR) model is adopted to effectively simulate the stochastic and non-stationary time sequences using a series of AR coefficients with a typical order. Artificial neural network (ANN) is utilized to distinguish the different force levels and game grades in the scenario of arm-wrestling. To advance the training speed and accurate rate of the motion pattern classification, back-propagation (BP) neural network based on adaptive learning rate algorithm (ALR) is introduced. The advantage of ALR algorithm compared with standard BP algorithm is confirmed by experiments

Application of Neural Network to Nonlinear Static Decoupling of Robot Wrist Force Sensor

The static coupling of wrist force sensor is a major influencing factor to its measuring precision. Aiming at resolving the disadvantages such as low decoupling precision of the traditional method, we put forward a nonlinear decoupling method based on neural network. The major idea applied is to use the BP network to realize the mapping from input to output of the sensor. Owing to BP networks good nonlinear mapping ability, the decoupling result can reach an arbitrary precision theoretically. The effectiveness of this method was verified in the calibration of wrist force sensor of a force sensing system for an underwater robot gripper. The decoupling results demonstrate the validation of neural network method

A novel thin six-dimensional wrist force/torque sensor with isotropy

Force sensors play a key role in the modern technological world, which can measure force information required for mechatronic systems used in automated manufacturing environments to function effectively and make intelligent decisions. Aim to decrease the additional torque originated from the height of the sensors, we design a thin six-dimensional force senor which is only 15mm high. The sensor has a simple geometry in the form of double E-type membranes connected with four thin rectangle slices. The sensor also possesses the advantages such as configurational simplicity, isotropic and high sensitivity.

Prediction of Human Elbow Torque from EMG Using SVM Based on AWR Information Acquisition Platform

In this paper a novel prediction method of elbow torque from EMG signal using SVM is proposed. How to model the relations between EMG signals and various kinematical aspects of the movement behavior is a difficult problem in the researches of neurophysiology and biomechanics. Traditional prediction methods include using neural networks to model the relations. However, these methods suffer from several problems, such as local minima, the difficulty of the selection of the model, etc. To address these problems, support vector machine is adopted to construct the nonlinear model. The efficiency of our proposed method is proved by experiment results.

Auto altitude holding of quadrotor UAVs with Kalman filter based vertical velocity estimation

Altitude regulation is a fundamental control problem of quadrotor UAVs (unmanned aerial vehicles) to ensure required performance for hovering and autonomous navigation. In this paper, under the assumption that the attitude inter loop has been well designed, the stability features of linear P+D+DD controller for altitude control are investigated. To handle the difficulty that the altitude velocity cannot be measured directly, a Kalman filter based ascending/decending velocity estimator using the measurements of acceleromter and barometer/sonser is adopted for velocity feedback. Given the overall dynamics model, P+D+DD controller design procedure and its stability analysis, the effectiveness of the algorithms is validated by field test using our quadrotor. The approaches can be directly extended to the translation velocity estimation and navigation control.

Design of a novel six-dimensional force/torque sensor and its calibration based on NN

This paper describes the design of a six-axis force/torque sensor, the purpose of which is to provide decoupled and accurate F/T information for the closed-loop control of the manipulator system. Firstly, the manipulator system and the adopted measuring principle are introduced. Then, a novel static component based on dual annulus diaphragms is presented. At last, the calibration and decoupling test based on Neural Network (NN) is carried out. The results of calibration test show superiority of the structure of the elastic component of the developed sensor and the improvement of the calibration method.