Shopping for food Colleague.


47 views
Uploaded on:
Description
GroZi venture (shopping for food right hand) Increase autonomy of individuals with low vision (exceptionally visually impaired) to perform shopping for food in a grocery store or store. ...
Transcripts
Slide 1

Shopping for food Assistant Carolina Galleguillos Pixel-bistro/June 2 2006

Slide 2

Description GroZi venture (shopping for food collaborator) Increase autonomy of individuals with low vision (exceptionally visually impaired) to perform shopping for food in a general store or store. Plan shopping list, strolling way to the store and shopping for food.

Slide 3

Motivation 1.3 million legitimately daze individuals in the U.S Grocery store are underselling to this business sector. Blind individuals are "high cost" clients. Advance examination on article acknowledgment for versatile mechanical autonomy with compelled processing assets.

Slide 4

Motivation Characteristics Grocery Store: Structured Environment (+). Controlled Lightening (+). Kept up by staff (+). Very much recorded (+). Individuals moving around paths (- ). Enormous measure of items (30K) (- ).

Slide 5

Motivation Possible existing arrangements: Seeing-eye canine prepared. RFID labels (passageway, rack, item). Standardized tag filtering (rack). Help of located aide/client administration. Retain store format. Home conveyance.

Slide 6

Why PC vision? Restricted capacity of pooches. RFID labels bring security concerns and overwhelming base. Eye security and mislabeling. Autonomy. Store design changes always. Self-rule.

Slide 7

Our Solution Develop a handheld gadget that performs visual article acknowledgment with haptic input. Profit of correlative assets (RFID, Barcode check, located aide) We are concentrating on the PC vision parts of this issue.

Slide 8

MoZi Box General reason ease versatile framework designed for PC vision applications. MoZi is a mix of the Mobile Vision System (MoVs) and ZigZag Finite memory : Compact Flash (CF) cards going from 256 MB to 4 GB. Processor speed: in the area of 60-400MHz Frame rate: enough depictions to cover the rack with some cover (as in all encompassing sewing) (15fps rather than 30fps?). Shading Calibration: Macbeth shading outline to align the shading space.

Slide 9

Use of the System Creating a Shopping List. Getting to the Grocery Store. Exploring the Store.

Slide 10

Shopping List Online Website: Website stores information and pictures of various items. Criticism from clients. Gives strolling way. Plan shopping list. Download data into Mozi Box.

Slide 11

On the way Separate undertaking. Mozi Box with GPS. Visual waypoints. Activity/Street sign perusing. Use notwithstanding stick and asking located spectators.

Slide 12

Inside the Store Finding walkway (OCR, RFID, inquire). Staying away from hindrances (stick). Discovering items (breadth of passageway, spot item, standardized identification check). Looking at (coupon and money).

Slide 13

Obtaining preparing information Online Images (Web). Gathering from MoZi box (in situ). Gathering from installed camera close to the standardized tag scanner (in situ). Known databases (COIL-100,ETH-80, etc.)(more research arranged) Synthetic cases. Dynamic learning.

Slide 14

Obtaining preparing information Active learning issue: Find UPC for the relating picture (marking). Semi-regulated. Feebly marked.

Slide 15

Obtaining preparing information Sunshine Store @ UCSD. Venue for pilot study. 4K things in stock. 1749 sq. ft. (assignable) We need to scale to a greater number of items (30K). No pastry shop or vegetables.

Slide 16

Object Recognition 2 sorts of acknowledgment (m:n, m<<n): Detection (of articles). Confirmation (objects distinguished are in that rundown). Calculations: SIFT, AdaBoost course, Multiclass Adaboost, Probabilistic Boosting tree, Color histogram coordinating, and so forth

Slide 17

Text Detection Standard OCR is unrealistic to be adequate. Low determination and twisting are primary issues. Perusing path signs, content on racks.

Slide 18

Considerations Occlusion and mess of items (brought about by individuals and shopping baskets). Different pictures of same rack to perform "opening fill-in". Can\'t fit prevailing plane to the front of item retires. Huge number of things.

Slide 19

Acknowledgments People (UCSD/Calit2): Serge Belongie John Miller Stephan Steinbach Michele Merler Tom Duerig Captions: Dennis Metz/D. Stein [ X. Chen and A. Yuille ]

Slide 20

Questions? Remarks?

Recommended
View more...