Data Science per il monitoraggio dell'assistenza sanitaria ...
Sistemi rinconfigurabili a basso consumo per il monitoraggio distribuito fine terzo... · Sistemi...
-
Upload
nguyenquynh -
Category
Documents
-
view
221 -
download
0
Transcript of Sistemi rinconfigurabili a basso consumo per il monitoraggio distribuito fine terzo... · Sistemi...
Sistemi rinconfigurabili a basso consumo per il monitoraggio distribuito
Tutor: prof. Luca Benini
Michele Magno
Outline
Wireless Video sensor NodeWireless Video sensor Network, WVNPower issueEnergy HarvestingPIR SensorMultomodal low cost systemApplication of video surveillanceResource manager for embedded computer
Wireless Video Sensor Network (WVSN)WSN with cameras for ambient monitoring
UnobtrusiveNo cables (remote locations)
Power availability issueBattery poweredSustainable operation
ProcessingWireless communication
Multimodal surveillance networksAugment informationReduce power consumption
Energy harvesting techniques to extend node lifetimeExtend nodes lifetime
Power issueEnergy consumption issue
Image acquisition and processingWireless transmission of recorded transmission
Several techniques can be used to extend network lifetime
Multimodal surveillanceEnergy harvestingLocal processing of imagesResource Manger
Energy efficient
Multimodal sensing
Energy harvesting
Local computation
Energy efficient Wireless Video Sensor NodeCombines three modalities
A Pyroelectric InfraRed (PIR) sensor provides a low power wake up signal. Sensor sensibility is modulated according to available energy and image contrast.
A PhotoVoltaic (PV) module is used to harvest energy from the environment.
Images are locally classified using Support Vector Machines (SVM) to detect relevant event and limit wireless transmission of undesired images
Wireless Video Sensor Node
Samrt Camera =Camera + intelligenceThe basis for new applications
Such as: detection, tracking, scene analysis
APPLICATIONS:• Automotive• Mobile Comm.• Surveillance• Consumer
Wireless Video Sensor Node –
(II)
Low costLow power consumptionLocal "intelligence" (on-board image
processing capability)A reconfigurable architecture to achieve
flexibilityWireless connectivity
Increase battery live
Wireless Video Sensor Node –
(III) Processing unit
Technology
Processor approach:DSP,
Media Processor,GPU,ARM
Programmable Logicapproach:
CPLD,FPGA,
Application
with
smart
camera People Detection
MICROCurrent Frame
Background Subtraction Winows Transfer Feature
ExtractionEmbedded Support
Vector Machine(ERSVM)
Background
FPGA
Video Processing Block
Application –
People detection IIARM9 – STR912
Three frame difference: background subtraction Region of Interest (ROI) detection (128x64 pixels)Feature extraction: average gray value over column and row (192 features)Classification: SVM-like hardware oriented algorithm
Sensor conditioning circuits: band pass filter (0.2-12Hz) + amplifier (1500x) + trigger generator
Low power: 1.5mWDetects changes in incident IR radiation
Moving bodiesPIR output dependencies
SpeedTemperatureDistance
Sensors PIR
+
+Out
Vdd
GND
Sensitive elements
Lens array Incident radiation
Absorbing structure
Architecture video sensor node + PIR
Typical use of PIR Sensor
VIDEO PROCESSING
Wireless Transceiver
2.4GHz
SLEEP MODE
Typical scenario
Application III –
abandoned /removed object detection with PIR Sensor
Active without video sensor
NO EVENTS!Sleep mode, only
PIR active
… …
EMPTY SCENE PIR->START
VIDEO PROCESSING
Application IV –
ZigBee WSN PIR- CAMERA
• Zibgee comunication• Scalable topology• Interaction Pir sensors and cameras motes
Application IV –
ZigBee WSN PIR- CAMERA
Power consumption measurementExtended battery life Node coopeation
Energy Harvesting & MPC ( ETH-
ZURICH)
• Model Predictive Control for rates and other parameters• Distributed MPC• Control the PIR sensitivity• Advance Multimodal video surveillance system
Resource Manager for embedded system (SCALOPES EU PRPJECT & STM)
Hw Design & Implementation
Thermal sensorPower IslandsClock gatingClock scaling
Sw DesignPorting of the Linux RM layer on SPEAr platformAdding RM feature to the IPs driver
SPEAR
ARCH-DRIVER
Normal Mode
Slow Mode
Doze Mode
Sleep Mode
Freq. of Cores
Freq of DDR
Freq of Perif
CPUIDLE CPUFREQACPI-DRIVER
RM-DRIVER
GOVERNORS
GENERIC INFRASTRUCTURE
SPECIFIC SPEAR GOVERNOS
USER LEVEL INTERFACES
APPLICATIONS/SURVEILLANCE APPLICATION
•/sys/devices/system/cpu/cpuidle
•/sys/devices/system/cpu/cpufreq
• Objective: A Linux-based RM infrastructure for the configurable SPEAR platform (multi-core processor + domain specific configurable accelerator)
CollaborationsPROJECTS
EU PROJECT: Scalopes ‘09-’10 SensactionAAL ‘08 – ‘09
National PROJECT:SUMMIT ‘07-’08PRIIN
Industrial STMicroelectronics ( Microcontroller, Vision, SCALOPES)Datalogic ( wireless applications, SUMMIT)
AccademicETH – Zurich ( 6 months intership) (MPC & Energy harvesting)EPFL (Andrea Acquaviva - reconfigurable architecture )University of Trento ( Support Vector Machine)PoliMI ( SCALOPES porjects)PoliTO ( SCALOPES porjects)
PublicationJournals:
Transactions on HiPEAC-1, Lecture Notes in Computer Science (LNCS), Springer-Verlag Berling Heidelberg New York Titolo: Exploration of Reconfiguration Strategies forEnvironmentally Powered DevicesAutori: Alex E. Susu, Michele Magno, Andrea Acquaviva, David Atienza, Giovanni De MicheliRivista: Journal of Real-Time Image Processing Titolo: A low-power wireless video sensornode for distributed object detectionAutori: Aliaksei Kerhet, Michele Magno, Francesco Leonardi, Andrea Boni and Luca Benini
Conferences:Congresso: Advanced Video and Signal based Surveillance Titolo: Distributed Video Surveillance Using Hardware-Friendly Sparse Large Margin Classifiers Autori: AliakseiKerhet, Francesco Leonardi, Andrea Boni, Paolo Lombardo, Michele Magno, Luca BeniniLuogo: Londra Data: Settembre 2007Congresso: 11th EUROMICRO Conference on Digital System Design Titolo: A Solar-powered Video Sensor Node for Energy Efficient Multimodal SurveillanceAutori: Magno, Michele; Brunelli, Davide; Zappi, Piero; Benini, LucaLuogo: ParmaData: Settembre 2008Congresso: Workshop on Multi-camera and Multi-modal Sensor Fusion Algorithms and Applications (M2SFA2) Titolo: Multi-modal Video Surveillance aided by Pyroelectric InfraredSensors Autori: Michele Magno; Federico Tombari; Davide Brunelli; Luigi Di Stefano; Luca Benini Luogo: Marsiglia Data: Ottobre 2008
Publication
(II)
Congresso: Sixth IEEE International Conference on Advanced Video and Signal BasedSurveillance Titolo: Multimodal Abandoned/Removed Object Detection for Low Power Video Surveillance SystemsAutori: Michele Magno; Federico Tombari; Davide Brunelli; Luigi Di Stefano; Luca BeniniLuogo: GenovaData: Settembre 2009Congresso: Third ACM/IEEE International Conference on Distributed Smart Cameras(ICDSC 2009) Titolo: Adaptive Power Control for Solar Harvesting Multimodal Wireless Smart Camera Autori: Michele Magno, Davide Brunelli, Lothar Thiele and Luca Benini Luogo: Como Data: Agosto 2009
POSTERS:Congresso: International Conference on Distributed Smart Cameras Titolo: A Low-PowerConfigurable Wireless Video Sensor Node for Distributed Vision Applications Autori: MAGNO M., L. BENINILuogo: Vienn aData: Settembre 2007Congresso: Fifth European conference on Wireless Sensor Networks, EWSN 2008 Titolo: A Self-powered Video Node Triggered by PIR Sensors Autori: M. Magno, D. Brunelli, P. Zappi and L. Benini Luogo: Bologna Data: Gennaio 2008Congresso: 6th European Conference on Wireless Sensor Networks Titolo: Detection ofabandoned/removed objects with a video sensor node aided by Infrared Sensor Autori: Michele Magno, Davide Brunelli, Luca Benini Luogo: Cork, Ireland Data: Febbraio 2009
Publication
(III) –
Work in progress
Congresso: The 9th ACM/IEEE International Conference on Information Processing in Sensor Networks (IPSN2010) Titolo Energy Efficient Cooperative Multimodal Ambient MonitoringAutori: Magno M., Zappi P., Brunelli D., L. BENINI STATUS: Submitted
TITOLO: Energy aware multimodal video surveillance embedded system. Autori: Magno M., Lanza A. Brunelli D., Di Stefano L., Benini L.
TITOLO: Resource manager for video surveillance embedded system. Autori: Magno M., Brunelli D., Benini L.
Looking for Transactions of Embedded computer and Embedded video and data signal processing
Conclusion and Future WorkEnergy efficient wireless video node
Multimodal surveillanceLocal preprocessing
Multi sensor node for distributed surveillanceVideo processing applicationData fusion for more energy efficiency
Energy Harvesting video sensor nodeResource managerFuture work
Augment information from low-power sensorsDistributed policies for collaborative surveillanceNew vision algorithms