ImageScape

From CENS Urban Sensing

[edit]

ImageScape Project

STAT 257 Project

For more information look at: http://docs.google.com/View?docid=dc574p8n_85p3pmz

Andrew Parker, Sasank Reddy, Josh Hyman

Introduction

Cell phones are becoming an ubiquitous, always on source of sensing in the urban environment. In particular, cell phones with cameras are starting to be used for lots of personal sensing applications. The cost, flexibility, and convenience of the platform are the main advantages of the platform.

What happens if a cell phone was worn and collected images at regular intervals throughout the day? The types of images have different attributes compared to traditional personal photography which are usually intentional and restrained in some way. Surveillance cameras are also intentional and situated (fixed). With cell phone cameras, the pictures are collected automatically and or mobile (auto-mobile).

Often time, these continuous picture capture applications exhibit redundant photography, many times are too dark, and exhibit other characteristics such as being too blurry.

Also, many images contain personal information, such as items like license plates, images of computer screens, faces, etc...

Problem Statement

How can we reduce noise associated with the images collected and help with the privacy problem? Or in another words, how can we selectively share the photography.

We plan to explore some simple filters for getting rid of images that are not useful, such as using color histogram analysis to get rid of dark images and redundant images, using blur detection methods by using such things as Haar features and edge analysis, and employing techniques to detect faces and text (Haar Classifiers).

Other Ideas

Could we employ the spatio-temporal locality to improve how we eliminate the noise related to the data? What happens when we have multiple cameras in the same area? How is privacy still protected in this case.

Progress

  • Created an initial version of ImageTagger. This will have to be refined a bit, but we are able to tag images with location and activity and save it on the phone.
  • Ordered 12 mini-SD cards for the phone (2 gig versions). This will be delievered by 01-24-06.
  • Asked Shaun Ahmadian to concentrate on images that are not related to Urban Sensing. Instead they will look into environmental photos (NestBox). The goal is to try to have a similar architecture for the image processing portion so that enviro-specific algorithms can be included later.

References

Rapid Object Detection Using a Boosted Cascade of Simple Features by Viola, P. and Jones, M.

Fast and Robust Classification using Asymmetric AdaBoost and a Detector Cascade by Viola, P. and Jones, M.

Detecting Faces in Images: A Survey by Yang, M.H. and Kriegman, DJ and Ahuja, N.

Face Recognition: A Literature Survey by Zhao, W. and Chellappa, R. and Phillips, PJ and Rosenfeld, A.

TODO


Sasank

  • Cluster visualization and interaction
  • Cell-id filtering
  • Map per image
  • Density map for selected images

Andrew

  • Add django code to SVN
  • Zip output via django
  • STAT236 Project
  • Think about mapping queries to URLs.
  • Think about filter stacks.


Josh

  • Clustering Eric's meadow images


Next week: Drag and Drop Convergence (Andrew and Sasank) KL stuff - HC, Drag and Drop Clusters (Andrew and Sasank) A map component (Sasank) Process control (all)

Deferred:

Blob count. Sensor base input/output. SVN - check out flex, django, apache configs, etc. Play the audio file. Spectrogram of the audio files.

Retrieved from "http://www.lecs.cs.ucla.edu/urban-sensing/index.php/ImageScape"