User talk:AnalysisDATA

A wearable fall detection system was proposed in [59] that could determine fall events by employing acceleration and orientation thresholds. The acceleration thresholds were obtained at the training phase from SVM, and the postural orientation thresholds were determined from the subject’s tilt angle. The system used Madgwick’s orientation filter for reducing magnetic distortion and gyroscope drift, resulting in high estimation accuracy. The IMU was placed on the waist and could communicate over Bluetooth. The system analyzed the RMS data obtained from the accelerometer and the orientation filter and could detect fall events using a threshold based algorithm. This allows implementing the algorithm for real-time applications in a low profile microprocessor. The algorithm was reported to achieve a high degree of accuracy and sensitivity. Table 4 presents a comparison of the key features and performance characteristics among the activity monitoring systems discussed above.

Proposition	Feature Extraction	Classification Method	Sensors	Sensor Placement	Com. Tech.	Detection	Accuracy	Power Req.
Activity and gait recognition system on a smartphone	Fixed set of features	Support Vector Machine (SVM), Bayes network, and Random Tree	Accelerometer is embedded in smartphone			Different walking speed	>99%.
In-home, fine-grained activity recognition multimodal wearable sensors	Fixed feature set	Conditional random field (CRF)	Smartphones’ (Samsung Galaxy S4) onboard sensors (accelerometer, gyroscope, barometer, temperature and, humidity sensor), along with Gimbal Bluetooth beacons	Waist, lower back, thigh, and wrist	USB	Walk and run indoors, use refrigerator, clean utensil, cook, sit and eat, use bathroom sink, move from indoor to outdoor, move from outdoor to indoor, walk upstairs, and walk downstairs, stand, lie on the bed, sit on the bed, lie on the floor, sit on the floor, lie on the sofa, sit on the sofa, and sit on the toilet	19 in-home activities with >80% accuracy
Wearable device based on a 9-DOF IMU	Fixed set of features		Accelerometer, gyroscope, and magnetometer	Limb or trunk	Bluetooth	Balance hazards, balance monitoring for fall prediction	High correlation	Streaming ~6 h Logging > 16 h
Algorithm development	Time-Frequncy domain analysis	Hidden Markov Model	3-axis accelerometer, 3-axis gyroscope	Chest	USB	Walking, running, ascending upstairs, descending downstairs and standing	~95%
A real-time, adaptive algorithm for gait-event detection			Two inertial and magnetic sensors ( 1 IMU = 1 accelerometer, 1 gyroscope)	External part of both shanks		Gait events: Initial Contact (IC), End Contact (EC) and Mid-Swing for both right and left leg while walking at three different speed	F1-scores 1(IC, EC), 0.998 (IC) and 0.944 (EC) for stroke subjects	F1-scores 1(IC, EC), 0.998 (IC) and 0.944 (EC) for stroke subjects
Recognition method for similar gait action	Inter-class relation Ship	Support vector machine, K-nearest neighbor	3 IMUs (each IMU: 1 tri-axis accelerometer,1 tri-axis gyro)	Fixed at the back, left, and right waist		Walking on flat ground, up/down stairs, and up/down slope	~93% average
Stochastic approximation framework	Fixed set of features	K–means and Gaussian Mixture Models	Accelerometer	Belt-like strap around the waist		3 intensity level of walking: 93.8%; 3 intensity level of running 95.6%
Power-aware feature selection for minimum processing energy	Minimum cost feature selection by using a redundancy graph	K-nearest neighbor	6 IMUs (each IMU has one three-axis accelerometer and a two-axis gyroscope)	Waist, right wrist, left wrist, right arm, left thigh, right ankle	BSN	Switching between stand and sit, sit and lie, bend to grasp, rising from bending, kneeling right, rising from kneeling, look back and return, turn clockwise, step forward and backward, jumping	30% energy savings with 96.7% accuracy
Parameter optimization strategy for phase-dependent locomotion mode recognition	Fixed set of features		2 IMUs, 2 pressure insoles (each having 4 pressure sensors)	IMUs on the shank and the shoe, pressure sensors insole		Walking, up/down stairs, and up/down slope, passive mode	88%–98%
Electronic insole for wireless monitoring of motor activities and shoe comfort	Fixed set of features		Humidity and temperature sensors, accelerometer and 4 pressure sensors	Insole	ZigBee	Foot accelerations, orientation in space, temperature and moisture data		10 h of data logging
Shoe-based activity monitoringsystem (smartshoe)	Fixed set of features	Support vector machine, multilayer perception (MLP)	Five pressure sensors (PS) and one 3-D accelerometer	PS on insole and accelerometer on heel of shoe		Sit, stand, walk, ascend stairs, descend stairs and cycling	99.8% ± 0.1% with MLP
A wearable device for monitoring daily use of the wrist and fingers	Fixed set of features	K-means	2 tri-axial magnetometers	Watch-like enclosure worn on the wrist and a small neodymium ring worn on the index finger	Finger and wrist movement	92%–98% with a 19%–28% STD	20.5 mA at 3.3 V
Combined kinematic models to estimate human joint angles	Unscented Kalman filter	Unscented Kalman filter	3 IMUs	Upper arm, forearm, and wrist	Shoulder internal/external rotation; flexion/extension of shoulder, elbow, and wrist, supination/pronation of forearm, wrist twist	Shoulder internal/external rotation; flexion/extension of shoulder, elbow, and wrist, supination/pronation of forearm, wrist twist	Average RMS angle error ~3°
Wearable device with automatic gait and balance analyzing algorithms for Alzheimer patients (AP)	Fixed set of features	Fixed set of features	3 IMUs (each IMU has a 3-d accelerometer, a uni-axial gyroscope, and a biaxial gyroscope	On feet for gait analysis on waist for balance analysis		Gait parameters and balance		30 mA at 3.7 V
IMU based fall Detection system	Madgwick orientation filter	Accelerometer, gyroscope, and magnetometer	Accelerometer, gyroscope, and magnetometer	Waist	Bluetooth	Backward fall, forward fall, lateral left fall, lateral right fall, syncope	Accuracy: 90.37%–100% Sensitivity: 80.74%–100%	15 mA–34 mA using 3.7 V