Presentation is loading. Please wait.

Presentation is loading. Please wait.

Panoptes: Low-Power, Scalable Video Sensor Networking Technologies Wu-chi Feng, Ed Kaiser, Brian Code, Mike Shea, Wu-chang Feng, Louis Bavoil Department.

Similar presentations


Presentation on theme: "Panoptes: Low-Power, Scalable Video Sensor Networking Technologies Wu-chi Feng, Ed Kaiser, Brian Code, Mike Shea, Wu-chang Feng, Louis Bavoil Department."— Presentation transcript:

1 Panoptes: Low-Power, Scalable Video Sensor Networking Technologies Wu-chi Feng, Ed Kaiser, Brian Code, Mike Shea, Wu-chang Feng, Louis Bavoil Department of Computer Science and Engineering OGI School of Science and Engineering at OHSU

2 www.cse.ogi.edu/sysl Motivation  Sensor networking technologies are great  Real-time in situ measurement of environments Habitat monitoring (UCLA) Columbia River forecasting (OGI) REINAS Monterey Bay system (UC Santa Cruz) Artic web cam (NOAA)  Video sensor networking technologies  Can add eyes to sensor data  Require significant computing and bandwidth resources beyond traditional sensor technologies

3 www.cse.ogi.edu/sysl Motivation  The applications  Environmental monitoring Example: Video sensor every ¼ mile along the entire Oregon coast  Health care delivery Example: Privacy ensuring elderly health care  Emergency response  Habitat monitoring  Surveillance and security  Robotics

4 www.cse.ogi.edu/sysl Motivation  Video sensor networking challenges  Low-power, power-aware video sensors PoE applications Environmental / autonomous deployment  Providing mechanisms that allow the sensor network to be tailored to specific applications “Programmability”  Managing information implosion (N  1) Buffering and adaptation  Making it easy to access both traditional scalar and video data within the sensor network

5 www.cse.ogi.edu/sysl The Panoptes Project at OGI  The goal:  Flexible, extensible middleware that supports massively scalable video-based sensor networks  Short term  Low-power, programmable, adaptable, video sensor for experimental testbed  Buffering and adaptation algorithms for sensor  Bringing together a large number of flows  Longer term  Integration of traditional low-power sensors with video sensors  Application-specific extensions

6 www.cse.ogi.edu/sysl The Rest of This Talk  The Panoptes platform  Hardware and software systems  Software architecture  Experimentation  A demonstration system  The Little Sister Sensor Networking Application  Conclusions and future work

7 www.cse.ogi.edu/sysl The Panoptes Platform  Picking a platform  Berkeley Motes  COTS web cameras  General embedded CPU platforms USB-based video 206 MHz Intel StrongArm Embedded Linux 802.11 wireless 320x240 video 22 fps software compressed ~5.5 Watts maximum

8 www.cse.ogi.edu/sysl The Panoptes Platform Video Sensor Architecture Buffering and Adaptation Supports disconnected or intermittent operation Priority mapping of streaming data elements Video 4 Linux Compression IPP-based Currently: JPEG, Diff JPEG, Cond. Replenishment Application-Specific Filtering Event-detection Time-elapsed images Computer vision Time Power Management

9 www.cse.ogi.edu/sysl Buffering and Adaptation  Sensor streaming is different than video streaming today  Live streaming Late data useless Data unknown a priori Limited use of buffering in adaptation  Video-on-demand streaming Just in time delivery All data known a priori Streaming can take advantage of known data Buffering useful How long to keep data in the sensor buffer? How do you prioritized data between new/old?  Sensor streaming Any data might be good Buffering can be used Some data unknown a priori Inverse multicast

10 www.cse.ogi.edu/sysl Experimentation  The USB bottleneck  Compression performance on Panoptes  Buffering and adaptation performance  Power measurements

11 www.cse.ogi.edu/sysl USB Capture Performance Image size USB Comp. Frame Rate% Sys CPU 160x120 029.644 129.7722 329.8816 320x240 04.883 128.7267 329.6845 640x480 0-- 114.1484 314.7378

12 www.cse.ogi.edu/sysl USB Capture Performance 6.9 Mbps Image size USB Comp. Frame Rate% Sys CPU 160x120 029.644 129.7722 329.8816 320x240 04.883 128.7267 329.6845 640x480 0-- 114.1484 314.7378

13 www.cse.ogi.edu/sysl USB Capture Performance 111 Mbps Image size USB Comp. Frame Rate% Sys CPU 160x120 029.644 129.7722 329.8816 320x240 04.883 128.7267 329.6845 640x480 0-- 114.1484 314.7378

14 www.cse.ogi.edu/sysl USB Capture Performance 27.6 Mbps Image size USB Comp. Frame Rate% Sys CPU 160x120 029.644 129.7722 329.8816 320x240 04.883 128.7267 329.6845 640x480 0-- 114.1484 314.7378

15 www.cse.ogi.edu/sysl Software Compression Performance Image size IPP (ms) ChenDCT (ms) 320x240 26.6573.69 640x480 105.84291.28 Image size IPP (ms) ChenDCT (ms) 320x240 19.4152.96 640x480 77.42211.42

16 www.cse.ogi.edu/sysl Capture / Compression Performance Image size IPP (ms) ChenDCT (ms) 320x240 29.2080.63 640x480 115.42319.71 Image size IPP (ms) ChenDCT (ms) 320x240 20.9557.31 640x480 83.95228.42

17 www.cse.ogi.edu/sysl Buffering and Adaptation

18 www.cse.ogi.edu/sysl Power Consumption Camera on (capturing) Camera standby Network connected Camera on/ net. connected All services running CPU loop System Idle Standby

19 www.cse.ogi.edu/sysl A Demonstration System  The Little Sister Sensor Networking Application NetworkNetwork Camera Manager(s ) Query Manager Stream Manager NetworkNetwork

20 www.cse.ogi.edu/sysl Future Work  Python-based experimentation  Power management  Developing a smaller (more stable) platform  Finding suitable radio technology to match applications  Making the access to video sensor data more useful  Integration with traditional sensor technologies  TinyDB for video sensors

21 www.cse.ogi.edu/sysl Conclusions  Low-power video sensor networking technologies  Video sensor software design Dynamically adaptable software architecture Disconnected or intermittent operation  More information  www.cse.ogi.edu/sysl

22 www.cse.ogi.edu/sysl

23

24 More information? http://www.cse.ogi.edu/sysl

25 www.cse.ogi.edu/sysl The Rest of This Talk  The Panoptes platform  Hardware and software systems  Software architecture  A demonstration system  The Little Sister Sensor Networking Application  Experimentation  System measurements  Buffering and adaptation  Power consumption  Conclusions and future work


Download ppt "Panoptes: Low-Power, Scalable Video Sensor Networking Technologies Wu-chi Feng, Ed Kaiser, Brian Code, Mike Shea, Wu-chang Feng, Louis Bavoil Department."

Similar presentations


Ads by Google