Body Area Networks for Human Motor Assessment

Transcription

1 Body Area Networks for Human Motor Assessment PhD candidate: Marco Benocci Supervisor: Prof. Luca Benini Micrel Lab DEIS - University of Bologna - Italy

2 BAN (Wireless) Body Area Network, consists of a set of mobile and compact intercommunicating sensors, either wearable or Figure Cou urtesy of Pedro Bran ndão Opera Lab implanted into the human body, which monitor vital body parameters and movements * * T. O'Donovan et al., A Context Aware Wireless Body Area Network (BAN), Inproceedings of the Pervasive Health Conference 2009.

3 BAN Challenges Sensors design System Devices Activities Remote Assistant Assessment large amount of data low complexity small form factor light weight power efficient easy to use reconfigurable monitor user to provide realtime feedback or assistance on emergency Emergency detection algorithm Radio and MAC Interoperability & Scalability information exchange across standards such as Bluetooth, ZigBee, UWB, Wi FI. the systems would ensure efficient migration across networks (Local Area, Internet) Security System and device level security limited access on private data accuracy data consistency

4 Application Driven Design Human Motor Assessment Monitor human posture and gait on: healthy to improve balance or to highline wrong activities subjects affected by gait disorder (PD, diabetics, ) Emergency detection algorithm: Feedback to the user (a/o remote alarm) on wrong posture/gait or danger of fall Sensors network design: wearable sensors (not implantable) form factor ~ 3x3x0.5cm signals: body kinetic and kinematic distribute processing (on board): > 5MIPS lifetime: >10h power budget: ~10mW number of nodes: < 10

5 Sensors Network Lower Limb 3D accelerometer (100Hz) Acc. 3D: ±2 g, 1.22mg Audio Stereo Modulation Trunk Haptic Feedback Frequency: Hz Up to 6 vibranting DC motors Heel IMU (140Hz) Acc. 3D: ±10 g, 2.54mg Gyro 3D: ±300 /s, /s Insole Hydrocells Array (140Hz) 24 hydrocells per foot Pressure: [0-625KPa] Fast stabilization: 20us

6 Postural Analysis 1/2 EU Projects: Collaboration: Reference work: M.Dozza et Al., Audio-Biofeedback for Balance Improvement: An Accelerometry-Based System, J. Biomedical Engineering, V.52, N.12, 2005 Frequency Intensity Balance Embedded platform constraints: form factor power budget computational power Node: 3D Accel ±2g, 1mg 8MIPS Fs = 600Hz/axis 10h BT 2.0 capability Gateway Xscale PXA MIPS 2h BT 2.0 capability

7 Postural Analysis 2/2 7 Subjects: 21-36yrs m, 57-75Kg Foam rubber (9cm thick) Simulating balance impairment Tasks: 1. standing on a firm surface (baseline condition) 2. sensory perturbation: standing the unstable support 3. sensory perturbation + ABF AP acceleration Low-pass filtered AP acceleration A. Baseline B. Sensory perturbation C. Sensory perturbation+abf 5 sec 20 mg TR

8 Different Feedback: Vibrotactile 4 different spatial-temporal patterns (foot/hand/trunk). Mechanoreceptors involved: Pacinian corpuscles, F Є [40-500] Hz: detect external acceleration or vibration Ruffini i endings, F Є [ ] Hz: provide buzz-like sensation Haptic Node: Eccentric mass vibrator motor Vibrating freq: 180Hz Max power consumption: 80mA PWM driven BT

9 Home Monitoring and Assisted Living Application 1/3: ZigBee optimization i EU Projects: SensactionAAL Collaborations: STM, ATOS

10 Home Monitoring and Assisted Living Application 2/3: ZigBee optimization i ZigBee protocol optimized in terms of throughput h t and power consumption for the fall detection scenario. System: 1 accelerometric node ZigBee stack 3.0 with duty cycling feature Throughput: 5Kbps Lifetime : ~80h

11 Home Monitoring and Assisted Living Application 3/3: fall detection ti algorithm Literature algorithms: single or double threshold, fall down speed Algorithm chosen: single-threshold h ld Feature: Sum Vector (SV) Tradeoff: detection VS false positive SV [m/s^ ^2] SV [m/s^2] Time[s] Time[s] Typology of fall Recognized Not recognized Forward fall 40 0 Lateral fall 7 0 Backward fall 10 0 Fall sliding against a 0 2 wall final position vertical (not in the recognition set) Fall sliding against a 3 0 wall, final position horizontal Falling out the bed 5 0 Total 65 2

12 Kinetic signals Re eal-time Posture CoP oscillation Gait CoP trajectory Gait FFT on single sensor Offl ine Gait Identify relevant foot area involved Gait Whole foot pressure map by interpolation Gait Gold Standard to identify kinematic sensors template

13 Foot pitch estimation during gait Real-time foot rotation detection Hp: motion only on the sagittal plane 1. Gyro integration gives Θ 2. Θ: acc. representation on external reference system (g static component deletion) az 3. gyroscope drift is compensation in z Z x midstance phase (az=0) X Θ EU Projects: Collaboration:

14 Wearable assistant for load monitoring 1/4 Intership: LONG TERM LOAD MONITORING DURING DAILY LIFE Automatically recognizing load during walking to make users conscious of their load-carrying habits and make necessary adjustments Features: Gait Interval (stance, swing,...) Biomechanical (trunk/hip inclination on sagittal and coronal plane,...) Energetic (energy efficiency) Features selected by filter approach (max entropy): Biomechanical

15 Wearable assistant for load monitoring 2/4 Windowing on right heel strike (t = T) [f1(t); ( ) f2(t); ; fn(t)] Ctemp(T) =F( f1(t); f2(t); ; fn(t) ) previous K estimes (K = 5): C(T) = [Ctemp(T); C(T -1); ; C(T- 1- K)]

16 Wearable assistant for load monitoring 3/4 ACTIVITIES DETECTED: N: normal walk SR, SL, SB (backpack) HR, HL (handbag) CLASSIFICATION RESULTS: 7 Subjects on 5 days Cross validation (F=4) on a single subject and on a single day High accuracy with different classifier S1 S2 S3 S4 S5 S6 S7 CRR Mean SVM NB KNN (K=3)

17 Wearable assistant for load monitoring 4/4 SIMPLIFY SENSORS SETUP Single accelerometer on the chest Classifier: KNN, N=3 CCR only Trunk sensor all sensors Day % 93.4% Day % 94.1% Day % 92.7% Day % 89.8% Day % 91.1%

18 Publications International Journal M. Benocci, C. Tacconi, E. Farella, L. Benini, L. Chiari, L. Vanzago, Accelerometer-based fall detection using optimized ZigBee data streaming, Microelectronics Journal (under review) International Conferences M.Benocci, M. Baechlin, E. Farella, D. Roggen, L. Benini, G. Throster, Wearable assistant for load monitoring: recognition of on body load placement from gait alterations, Pervasive Health 2010, (under review) M. Benocci, E. Farella, L. Benini: Optimizing ZigBee for data streaming in body-area bio-feedback applications. Proc. of the 3rd IEEE International Workshop on Advances in Sensors and Interfaces (IWASI- 2009), Bologna, Italy M. Benocci, L. Rocchi, E. Farella, L.Chiari, L. Benini: Wearable system for foot kinematics and pressure measurements during gait. Proc. of the 19th International Conference on Posture and gate 2009 (ISPGR- 2009), Bologna, Italy M. Benocci, D. Brunelli, E. Farella, L. Benini: A body area network with vibrotactile actuation. Proc. of the 19th International Conference on Posture and gate 2009 (ISPGR-2009), Bologna, Italy M. Benocci, L. Rocchi, E. Farella, L. Chiari, L. Benini: A Wireless System for Gait and Posture Analysis Based on Pressure Insoles and Inertial Measurement Units. Proc. of the 3rd International Conference on Pervasive Computing Technologies for Healthcare 2009 (PERVASIVE HEALTH-2009), London, UK L. Rocchi, M. Benocci, E. Farella, L. Benini, L. Chiari: Validation of a Wireless Portable Biofeedback System for Balance Control: Preliminary Results. Proc. of the 2nd International Conference on Pervasive Computing Technologies for Healthcare 2008 (PERVASIVE HEALTH-2008), Tampere, Finland L. Rocchi, E. Farella, M. Benocci, E. Santarmou, L. Benini, L. Chiari: Modular Architecture for a Wireless Bio- Feedback System for Balance Control. Proc. of the International Conference on Pervasive Health 2007 (PHealth-2007), Porto Carras, Greece

19 Conclusion In summary Design of WBAN low-power devices (mw) for motor analysis and biofeedback provisioning sensors explored: 6DOF inertial node (acc. + gyro) Baropodometric Insoles (capture of feet pressure map) Haptic feedback node Developed algorithm to: improve user posture (healthy/pd) provide real-time biofeedback (audio/haptic) detect gait phases recognize user on walking alterations due to load carriage Looking forward Extend distributed processing defining a multistage activities recognition. Embed classification directly on the nodes coping with constrained resources.