Hello, I'm Sekine of the digital business promotion office.
From this will include a technically inclined stories from intra-mart BIORA booth.
First first is a OCR and IoT (LED)!
Flow of processing at the reception
OCR and IoT (LED) will be technology is used at the time of reception.
Reception for BIORA booth, and it was and flow of processes such as the following.
- Customers I presented business cards. Photographed in an tablet
- Captured business card image character string in IM-OCR
- OCR result to morphological analysis, company name, job title, the name
- Generation machine learning data such as to select the taste of curry taste based on the acquired information
- Performed vacant confirmation, seat guide (while bashing LED is lit API, screen display)
- LED of the seat which is the guide target lit
Overview implementation of the reception function
1. Customers I presented business cards. Photographed in an tablet
Is made reception screen in intra-mart.
Available radio button that can be selected number of people and the camera starts up for file tag, a submit button (simple!)
2. Character string of captured business card image passed to the IM-OCR
Processing when the tapped "transmission" buttons on the screen is implemented.
※In IM-OCR, but where the text in OCR can be set as "block".
This time I block specify the entire business card.
3. OCR result to morphological analysis, company name, job title, the name
The morphological analysis data character string in IM-OCR.
Classification from the analytical result of "customer" "post" "person and family name" "person/name" have been used as respectively, company name, job title, the name.
Analysis results will be as follows:
Manager nouns, proper noun, Post, *, *, *, Department Manager, *, *
Aoyagi nouns, proper noun, people name, LastName, First *, *, Aoyagi, Aoyagi, *
Tatsumi nouns, proper noun, people name, name, *, *, Tatsumi, Tatsumi, *
NTT DATA INTRAMART CORPORATION nouns, proper nouns, customers, *, *, *, NTT DATA INTRAMART CORPORATION, *, *
〒 symbol, general,*,*,*,*,〒, yuubingbangou, yuewingbangou
107 Noun several,*,*,*,*,*
- Symbol, general,*,*,*,*,*
0052 Noun several,*,*,*,*,*
Akasaka, Minato Ward, Tokyo nouns, proper nouns, region, and general, *, *, Akasaka, Minato Ward, Tokyo, Tokyo tominatikakasaka, tokyoot minatikakasaka
* A twist here
"person and family name" Regard, since pattern that is recognized as a "name" exists
"person/name" When the previous word recognized as "name" additional processing that you want to replace with "person and family name", and so on.
Reading the business card of Mr. Takamatsu, it is recognized as a Takamatsu (city) in Kagawa Prefecture, so avoid this!
Incidentally, analysis percentages of correct answers EWS current day is no longer possible to or less.
"company name" Analysis Accuracy: 72.0%
"post" Analysis Accuracy: 92.5%
"family name" Analysis Accuracy: 84.9%
"name" Analysis Accuracy: 77.4%
※Costs were calculated by comparing actually captured business card and the OCR result information.
※Business card image or OCR result information, so please don't worry is deleted the correct answer ratio calculation after.
For "family name" read by the here I have been using an in AI voice chat
Of about 15% will be the fact that called mistakes in your name has occurred.
I'm very sorry I….
4. Generation machine learning data such as to select the taste of curry taste based on the post that was obtained
Flavored with curry, but we also implemented process to be estimated from statistical data based on the acquired information.
Implementation to select rare materials to other, fell under certain conditions.
Whether the taste of curry is to your liking, within a conversation of AI voice chat
To hear comments of customers.
Was summarized impressions, it will be as below.
"is wrong": 21.3%
※Received voice of "perfect!" "was good dry…." "I don't know eat the" "WOOOW" etc.!
5. Performed vacant confirmation, seat guide (while bashing LED is lit API, screen display)
And processing of as is completed, and the search a vacant seat, booth information may be displayed for the OCR screen of a while ago.
6. LED of the seat which is the guide target lit
LED has been using the full color LED tape.
Cut for each 2.5 m by solder connected to AC power and Wi-Fi module microcomputer!
Microcomputer is based on the esp-wroom-02
Are implemented control the feeling of spinning based on the image Skytree.
This article to here.
Next to the main I am going to write my article Roomba.