eric minner james pittman. outline project statement / motivation concept overview quick computer...

9
Eric Minner & James Pittman

Upload: roxanne-young

Post on 18-Jan-2018

217 views

Category:

Documents


0 download

DESCRIPTION

Project Description Participatory sensing often relies on receiving a great number of varied inputs to provide a useful service. The goal of the project is to design a system concept that will find novel images in the world. This system needs to have a way to help encourage a stream of interesting and novel shots.

TRANSCRIPT

Page 1: Eric Minner  James Pittman. Outline Project Statement / Motivation Concept overview Quick computer vision overview Demo Lessons Learned Future Work

Eric Minner & James Pittman

Page 2: Eric Minner  James Pittman. Outline Project Statement / Motivation Concept overview Quick computer vision overview Demo Lessons Learned Future Work

Outline• Project Statement / Motivation• Concept overview• Quick computer vision overview• Demo• Lessons Learned• Future Work

Page 3: Eric Minner  James Pittman. Outline Project Statement / Motivation Concept overview Quick computer vision overview Demo Lessons Learned Future Work

Project Description• Participatory sensing often relies on receiving

a great number of varied inputs to provide a useful service.

• The goal of the project is to design a system concept that will find novel images in the world.

• This system needs to have a way to help encourage a stream of interesting and novel shots.

Page 4: Eric Minner  James Pittman. Outline Project Statement / Motivation Concept overview Quick computer vision overview Demo Lessons Learned Future Work

Concept Overview• Leverage existing mobile phone capture and

image transfer protocols to allow users to submit images.

• Setup a backend server using apache, and a website to host “novel photo of the day”

• Use a computer vision algorithm to parse the submitted images and create a ranking scheme to encourage further participation for users to submit images

Page 5: Eric Minner  James Pittman. Outline Project Statement / Motivation Concept overview Quick computer vision overview Demo Lessons Learned Future Work

Concept Diagram

Camera capture w/EMAIL or

other image transfer

protocol

Mobile Phone or Emulator

Image Databas

e

Apache Server

HTML & PHP source for

“picture of the day” website

Backend Java Applicationfor image processing

and ranking

WWW

Backend PC

GMAILServer

Email Retrieval Applicatio

n(Outlook)

Image Metric Files

Page 6: Eric Minner  James Pittman. Outline Project Statement / Motivation Concept overview Quick computer vision overview Demo Lessons Learned Future Work

Wait for Image

Image Received

from Email

Java App detects

new Image

Extract Image

Features

Compare against

database

Accumulate

Similarity measure

Generate new image

IMF

Update database

IMFs

Add to database

Update Database Hierarchy

Update Webpage

Has the novel image

changed?

N

Y

Page 7: Eric Minner  James Pittman. Outline Project Statement / Motivation Concept overview Quick computer vision overview Demo Lessons Learned Future Work

Demo

Page 8: Eric Minner  James Pittman. Outline Project Statement / Motivation Concept overview Quick computer vision overview Demo Lessons Learned Future Work

Lessons Learned1. Choosing non subjective features for images is

very hard. When you automatically rank images based on a feature type you won’t always get the result that visually you think you should

2. Large images (1MB or larger) take forever to send to the server. We need something to handle memory size scaling (the iPhone has this built in, but not all devices do)

3. Resolution scaling (for proper display on the site) was an issue and currently we just force all the images to the same size

Page 9: Eric Minner  James Pittman. Outline Project Statement / Motivation Concept overview Quick computer vision overview Demo Lessons Learned Future Work

Future Work1. Integrate the image capture and submission

into a mobile phone application2. Improve image scoring / feedback process to

users by having the system return a ranking directly to the user either via email/SMS or as part of the application

3. Adapt system to work with other types of participatory sensing (such as: fuel prices or carbon footprint tracking)

4. Image scaling / compression when received by email