locaf detecting real world states with lousy wireless cameras n.
Download
Skip this Video
Loading SlideShow in 5 Seconds..
LoCaF : Detecting Real-World States with Lousy Wireless Cameras PowerPoint Presentation
Download Presentation
LoCaF : Detecting Real-World States with Lousy Wireless Cameras

Loading in 2 Seconds...

play fullscreen
1 / 28

LoCaF : Detecting Real-World States with Lousy Wireless Cameras - PowerPoint PPT Presentation


  • 59 Views
  • Uploaded on

LoCaF : Detecting Real-World States with Lousy Wireless Cameras. Benjamin Meyer, Richard Mietz , Kay Römer. Structure. Introduction Motivation Challenges System Architecture Evaluation. Motivation. SFpark project: http://sfpark.org/. Towards the Internet of Things

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha
Download Presentation

PowerPoint Slideshow about 'LoCaF : Detecting Real-World States with Lousy Wireless Cameras' - karl


An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
locaf detecting real world states with lousy wireless cameras
LoCaF:Detecting Real-World States with Lousy Wireless Cameras

Benjamin Meyer, Richard Mietz, Kay Römer

slide2

Structure

  • Introduction
    • Motivation
    • Challenges
  • System Architecture
  • Evaluation
slide3

Motivation

SFpark project: http://sfpark.org/

  • Towards the Internet of Things
    • High-level state of things on the internet
    • Scalar/specialized sensors are often limited to one scenario
    • Cameras are more flexible
slide4

Low-cost hardware

  • Sensor nodes
    • Constrained resources
  • Low-cost cameras
    • Low resolution
    • Poor image quality
    • Low frame rate
  • Processing is shifted to the gateway
slide5

Scenarios

Occupancy of a room

Free seats in a room

Individual occupancy of parking spots

Picture

Objects to detect

People

People

Cars

States

Free/occupied

Number of persons

Free/occupied for each parking spot

Challenges

Possibly lots of movement

Possibly lots of movement

Outdoor  Changing lighting conditions

Flexible Framework to infer and publish states for divers scenarios

slide6

HTML

System Architecture: Overview

RDF

Tweet

SQL

Customizable workflow

0

  • Image capture
  • Compression
  • Wireless transmission
  • Image processing
  • Regions of interest
  • Enhancing filters
  • Object detection
  • Face detection
  • Mobile object detection
  • State inference
  • Rule-based language
  • State publication
  • Text templates
  • Different media
slide7

HTML

System Architecture: Sensor Node

RDF

Tweet

SQL

Customizable workflow

0

  • Image capture
  • Compression
  • Wireless transmission
  • Image processing
  • Regions of interest
  • Enhancing filters
  • Object detection
  • Face detection
  • Mobile object detection
  • State inference
  • Rule-based language
  • State publication
  • Text templates
  • Different media
  • Camera equipped sensor node
  • Two capture modes
    • Time-triggered
    • Event-triggered (by PIR)
  • JPEG-compression in hardware
  • Fragmented transmission to gateway
slide8

HTML

INSTITUTE OF COMPUTER ENGINEERING

System Architecture: Processing

RDF

Tweet

  • Image processing
  • Regions of interest
  • Enhancing filters

SQL

Parking spot a

Parking spot b

Customizable workflow

0

  • Image capture
  • Compression
  • Wireless transmission
  • Image processing
  • Regions of interest
  • Enhancing filters
  • Object detection
  • Face detection
  • Mobile object detection
  • State inference
  • Rule-based language
  • State publication
  • Text templates
  • Different media

Region selection

Lighting compensation

Texture enhancement

Contrast enhancement

Orchestration and parameterization of enhancements

slide9

INSTITUTE OF COMPUTER ENGINEERING

System Architecture: Processing

  • Object detection
  • Face detection
  • Mobile object detection
  • Image processing
  • Regions of interest
  • Enhancing filters
  • Object detection
  • Face detection
  • Mobile object detection
  • State inference
  • Rule-based language
  • Face detection
  • Adaptive background subtraction
    • Classification into fore- and background
    • Can adapt to small changes
  • Blob detection
    • Each blob is an object

Number of & area covered by objects

slide10

HTML

INSTITUTE OF COMPUTER ENGINEERING

System Architecture: Processing

RDF

Tweet

  • State inference
  • Rule-based language

SQL

count:map:0:1:free

count:map:1:-1:occupied

State-based

free

count:map:0:1:All seats free

count:map:10:45:Enough seats

count:map:45:70:Almost full

count:map:70:-1:No seats left

Customizable workflow

80% coverage

80% coverage

0

  • Image capture
  • Compression
  • Wireless transmission
  • Image processing
  • Regions of interest
  • Enhancing filters
  • Object detection
  • Face detection
  • Mobile object detection
  • State inference
  • Rule-based language
  • State publication
  • Text templates
  • Different media

occupied

area:switch:free:80:occupied

area:switch:occupied:80:free

area:map:0:80:free

area:map:80:100:occupied

count:map:0:1:free

count:map:1:-1:occupied

Event-based

Rule-based state inference

slide11

HTML

System Architecture: Publishing

RDF

Tweet

SQL

Customizable workflow

0

  • Image capture
  • Compression
  • Wireless transmission
  • Image processing
  • Regions of interest
  • Enhancing filters
  • Object detection
  • Face detection
  • Mobile object detection
  • State inference
  • Rule-based language
  • State publication
  • Text templates
  • Different media
  • Every text format (HTML, RDF, TXT, …)
  • Template-based
  • Publishing via
    • FTP
    • Twitter
    • SQL-Database
slide12

Evaluation Setup

  • Camera in front of lecture hall during lecture
  • Estimate number of students
  • Also looking at binary state (free/occupied)
  • One region, background subtraction & no filter
  • Three phases:
    • Beginning: Entering persons in dribs and drabs
    • During: Not many movements
    • End: Abrupt leaving of students
slide13

Evaluation: Under- and Overestimation

  • Underestimation
    • Several persons identified as one
    • Persons not recognized because of no movement
  • Overestimation
    • Legs recognized as individual
slide14

Evaluation: Entry phase

OE: 130%

UE: 70%

  • Avg: 48%
  • Binary state always correct
slide15

Evaluation: Lecture phase

UE: 105%

  • Avg: 54%
  • Binary state always correct
slide16

Evaluation: Exit phase

UE: 222%

OE: ∞

  • Avg: 95%
  • Binary state not correct for picture 11-13
slide17

Evaluation: Entry phase revisited

Image filters can significantly change the estimation

slide18

Evaluation: Entry phase revisited

Parameters can significantly change the estimation

Improved avg error: 12%

slide19

Conclusion

Flexible framework

Use of cameras to be applicable in divers scenarios

Fully customizable by the user in each step

Accuracy quite high

slide20

Questions?

Thank you for your attention.

Time for questions.

slide21

Setup

Camera node

Gateway

Netbook with software

slide28

Evaluation: Parking Spot Scenario

area:switch:free:80:occupied

area:switch:occupied:80:free

Select single spot

State switches from free to occupied when car enters (b) and c))

State will switch back when car leaves