CENS Data Access

From CSL Wiki

Descriptions of available data sets and the information of interest.

Contents 1 Dietsense Images 1.1 High-level interests 1.2 DietSense Images 1.3 Information of interests 2 Nestbox Images 2.1 Wired 2.2 Wireless 2.3 Information of interest 3 Plant Phenology 3.1 Cameras 3.2 Information of Interest 3.3 Current Status

Dietsense Images

Improved dietary intake monitoring with participatory sensing using cellular phones. The purpose of this study is collecting information about an individual's daily dietary intake and food choice behaviour by logging daily diet through image capture.

High-level interests

Provide efficient mechanisms to navigate, annotate, filter, and analyse the collected data (images), including the capability to export reports to common statistical software packages. With these tools, participants will be able to directly configure, review and audit all automatic data collection features in the system.

DietSense Images

1. Pilot1: Energetics Pilot (Supporting Public Health Dept Research)
   1. Data is not accessible/sharable at this point
2. Pilot2: CENS internal pilot
   1. Duration: A single day pilot (Aug 14th, 2007)
   2. Data source: 14 CENS faculty/staff/student participants (agreed to share their images)
   3. Data collection rate: image/10 seconds
   4. Image Gallery is available for reviewing images here (Contact Deborah Estrin for access)
   5. For more information here

Information of interests

Goal: Participants need to be able to quickly audit and annotate images collected of their diet and authorize them for upload.

1. Filtering
   1. Poor quality images - Blurry, over/under exposed, noisy, etc.
   2. Redundant images - Repeating images (by clustering similar images)
2. Eating episode recognition? (possible future work)
   1. Pilot1: Energetics Pilot assumes the participants will collect images only when they eat.
   2. Pilot2: CENS Pilot collected images all day

• Current status

1. Implemented a web service that takes images as an input and provides four meta data of each images (mean_intensity, std_dev, blurriness, edge count)
2. Filters images based on a threshold that we choose manually by simply looking through the images
3. A CENS summer intern has written a report of her implementation - [Haleh2007]

Nestbox Images

Images of nestboxes have been taken and stored over several years with a wired NTSC camera since 2004. Various data sets include:

Wired

• Wired cameras: 480x704 images at 15 min intervals.
  • JR website archives/GUI: Example
  • JR Image Sets
• Wired cameras: 480x704 images (continuous ~ 1fps)
  • Vision/CENS Lab servers: _link goes here_

Wireless

• Wireless cameras w/o IR filter: 128x128 images at 15 min intervals.
  • CENS lab servers: Deployment 2006
    • One Set
  • Web GUI browsing: Deployment 2006 GUI
• Wireless cameras w/ IR filters: 280x240 images at ~8 min intervals.
  • CENS lab servers:
    • Deployment 2007 : Location 1
    • Deployment 2007 : Location 2
    • Deployment 2007 : Location 3
    • Deployment 2007 : Location 4
  • Web GUI browsing: Deployment 2007 GUI

Information of interest

• Time and duration of nesting stages: nest building, egg laying, incubation, hatching, raising of youngs
• Presence/Absence of bird, duration of visits by bird
• Number of eggs laid, number hatched, number fledged
• Feeding of chicks by mother
• Bird Species
• Feeding patterns

Plant Phenology

Images of plant phenology have been taken and stored over periods from about one year (tower cams at JR) to about 4 years (mosscam). Data sets include:

Cameras

1. Tower Cams
   • Targeted images of annual plants
   • Patchwork-panoramas of areas where annual plants may appear
2. NIMS mobile camera
   • Targeted images of Rhododendron
   • Targeted images of Bracken Fern
   • Zoomed-out images of areas of Bracken Fern
3. MossCam
   • Four years of 15-minute interval images
   • MossCam

Information of Interest

1. Flowering
   • Timing (annuals and perennials)
   • Number of flowers per species
2. Leafing
   • Timing (annuals and perennials)
   • Leaf area
3. Activity
   • Mosscam-related greenness associated with increased capacity for photosynthesis
4. PAR
   • Light reflected from leaf surfaces might be modeled from images, resulting in large-area estimates of received light for photosynthesis

Current Status

1. Browser and tools for color tasks and sorting of images has been created, assisting greatly in the analysis of images. Image Tool
2. Flowering
   • Hue-based subtraction of background for detecting color areas associated with flowers. Blur and then blob counting estimates flower abundance.
   • SIFT with default values is not effective for detecting certain flowers
     • Need work with SIFT or with template matching or other techniques to detect small flowers in complex background
3. Leafing
   • Hue-based subtraction of background for detecting color areas associated with leaves. Numbers of pixels associated with hue is related to leaf area.
     • Need ground-based measurements of more species leaf areas to correlate with images.
     • Need alternative leaf detection methods
4. Activity
   • Color ratio method is not great for detecting green periods.
   • Have ancillary data on temperature and light for better estimates of photosynthesis from images
5. PAR
   • Have multiple exposure and shutter speed images of ferns for light estimation
   • Need to make correlations between PAR and reflected light for times of day.