Indicon 2013 , Mumbai, 13-15 December 2013, Paper ID 1084 Track 4.1 Signal Processing & VLSI (Biomedical Systems & Signal Processing ) Sunday, 15-12-2013, 1540 – 1710. A Wearable Inertial Sensing Device for Fall Detection and Motion Tracking . Praveen Kumar - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Indicon2013, Mumbai, 13-15 December 2013, Paper ID 1084 Track 4.1 Signal Processing & VLSI (Biomedical Systems & Signal Processing )Sunday, 15-12-2013, 1540 – 1710
IIT Bombay
Praveen KumarPrem C. Pandey
erpraveen @ iitb.ac.in, pcpandey @ ee.iitb.ac.in
A Wearable Inertial Sensing Device for Fall Detection and Motion Tracking
• Low-cost, compact, & free from interference problems.• No restrictions on the movement space.
Observations based on the literature• Only accelerometer or only gyroscope: good results for restricted movement in specific
directions.
Multiple sensors: recognition of a larger types of activities, better accuracy.• System with sensors on multiple body parts for tracking relative movement of different body
parts.• System for fall detection: head, waist, trunk, and thigh found to be good sensor placement
locations, wrist found to be unsuitable.• Multiple signal fusion & fuzzy inference systems: enhanced accuracy but not well suited for real-
time applications. • Threshold based fall detection: well suited for real-time fall detection but lower accuracy.
2. HARDWARE DESIGNDesign objectiveContinuous acquisition of acceleration & angular velocity data: settable sampling frequency: 100 Hz or higher for gait monitoring and fall detection, < 20 hz for actigtraphy.Processing capacity for real-time fall detection.
Wireless connectivity: operation control, data transfer, fusion of data from multiple devicesInternal memory: data recording Compact & wearable: single supply operation with low power consumption, no switches & connectors.
ComponentsMEMS-based sensor with integrated tri-axial accelerometer & gyroscope; Microcontroller; Flash memory; Serially interfaced Bluetooth module; Regulator
Microcontroller16-bit microcontroller: Microchip PIC24F64GB004 (44 pin) 35 I/O pins, Two SPI, two I2C, two UART, one USB 64 KB program memory, 8 KB RAM,. Internal clock of 8 MHz FRC with fCY of 4 MHz Vdd: 2 – 3.6 V, Idd: 2.9 mA (at 4 MIPS)
Memory64-Mb serial dual I/O flash memory: Microchip SST25VF064C Nonvolatile memory for recording more than 12 hours of data for actigraphy;
Burst mode data transfer to save processor time for real-time fall detection and data transfer from multiple modules in a time multiplexed manner
Bluetooth ModuleSerially interfaced Bluetooth module: Roving Networks RN-42Range: 20 m range Data rate: 240 kbps in slave modeVdd: 3.3 V, Idd: 3 mA (connected) & 30 mA (data transfer)
PowerMCP 1802 LDO regulator: 3.3 V output for 3.5 – 12 V input, with max current of 300 mA.
Sample-by-sample data acquisitionRead the 6-axis sensor data at each sampling interval; save the data in internal 252 bytes buffer. If internal buffer is full, write 252 byte- data to the memory using page program
Burst mode data acquisitionRead 1024 bytes from FIFO at each interrupt; write to flash using page program; check for IRQ from UART and service it if needed.
PC based GUI for operation control & data transfer through Bluetooth
Test setup: Control Moment Gyroscope Model 750 (Educational Control Products)
Central platform with two outer rings Encoders to record the angles of rotation
using a PC Brakes for fixing angular positions
•Testing Device mounted on central platform Movements of platform or the rings Simultaneous recording of the sensor outputs by the device & encoder outputs using PC
4. REAL-TIME FALL DETECTION Observations from the accelerometer recordings
Fall: Large variation from the mean value for a certain duration and in a certain direction.
Multiple direction decomposition of accelerometer output and thresholding can help in improving sensitivity & specificity of the detection, without using gyroscope outputs.
Real-time fall detection method: Thresholding & duration window on 7 directional components
Components: Three axial components of the acceleration, magnitudes of the acceleration in three orthogonal planes, and the magnitude in the three-dimensional space