Author Topic: how to recognise overlapping images  (Read 165 times)

yotamarker

  • *******
  • Starship Trooper
  • Posts: 400
how to recognise overlapping images
« on: March 18, 2017, 07:28:31 AM »
how to recognise overlapping images ?

korrelan

  • *
  • Replicant
  • ********
  • Posts: 527
  • Look into my eyes! WOAH!
    • Main YouTube
Re: how to recognise overlapping images
« Reply #1 on: March 18, 2017, 10:39:51 AM »
Quote
how to recognise overlapping images ?

Do you mean for depth perception?

There a few ways to get depth from a machine vision system.  You could use something like a digital rangefinder or Kinect.

You can use the physical offsets from stereo cameras… adjust the physical angles of the cameras until you get an overlap of the image… the angles corresponds to a depth reading.

Our eyes give stereo images; the eyes can move horizontally in unison within their sockets but never vertically independently (or shouldn't do lol).  The idea is to select a small piece of image from one view and seek it horizontally at a similar location in the second image; the offset or horizontal shift equates to the depth.

I’ve just wrote some simple code to demonstrate how to obtain depth information for stereo images.  I’ve slowed it down so you can see how it maps one image against the other. Darker grey maps to deeper depth; you wouldn’t normally need to convert the whole image just a matrix of points… depending on your requirement.



Read the RGB values of small area of image (say 2X10 pixels, experiment with various sizes) into an array.

Then in the other image search at the same vertical position but slightly to the left… move the search to the right checking for a match as you go.

Depending on the image quality you might need RGB offsets for the matching, ie

.offset=10
.sred=original red from left image
.ored=red from the image we are searching

.If sred >= ored - offset and sred<= ored + offset then red_match =true

.same for blue/ green channels

The position difference when you find your match is the depth.

 :)
« Last Edit: March 18, 2017, 12:02:37 PM by korrelan »
It thunk... therefore it is!

yotamarker

  • *******
  • Starship Trooper
  • Posts: 400
Re: how to recognise overlapping images
« Reply #2 on: March 18, 2017, 02:00:11 PM »
I meant like in cursive OCR or like one car in front of half of another car.

keghn

  • *
  • Starship Trooper
  • *******
  • Posts: 302
Re: how to recognise overlapping images
« Reply #3 on: March 18, 2017, 02:25:53 PM »
 A image with over lapping object in video is done taking information from other frames when there was no interactions.
 It about tracking object into the chaos so they are not lost.

yotamarker

  • *******
  • Starship Trooper
  • Posts: 400
Re: how to recognise overlapping images
« Reply #4 on: March 18, 2017, 03:08:27 PM »
yeah I also saw that working but not for the cursive OCR case cause you don't
have enough detail, therefore it needs an extra solution

korrelan

  • *
  • Replicant
  • ********
  • Posts: 527
  • Look into my eyes! WOAH!
    • Main YouTube
Re: how to recognise overlapping images
« Reply #5 on: March 18, 2017, 04:25:14 PM »
So you are looking for robust system to read hand written cursive text?

Quote
you don't have enough detail

Are you using a scanner or camera to obtain the images?

Why do you think object occlusion/ recognition is relevant to OCR?

Exactly what are you trying to achieve?

 :)
It thunk... therefore it is!

yotamarker

  • *******
  • Starship Trooper
  • Posts: 400
Re: how to recognise overlapping images
« Reply #6 on: March 18, 2017, 05:31:29 PM »
using a camera;
//trying to achieve image recog

reconizing an object is one thing but :
if it is overlapping like 2 dolls standing and touching
or
two char's written in cursive

the recog fails

keghn

  • *
  • Starship Trooper
  • *******
  • Posts: 302

 

Welcome

Please login or register.



Login with username, password and session length
Choreographing automated cars could save time, money and lives
by Tyler (Robotics News)
Today at 04:49:29 AM
Three years on: An update from Leka, Robot Launch winner
by Tyler (Robotics News)
March 26, 2017, 10:49:19 PM
The Drone Center’s Weekly Roundup: 3/13/17
by Tyler (Robotics News)
March 26, 2017, 04:49:50 PM
Intel to acquire Mobileye for $15.3 billion
by Tyler (Robotics News)
March 26, 2017, 10:48:12 AM
Developing ROS programs for the Sphero robot
by Tyler (Robotics News)
March 26, 2017, 04:48:29 AM
Split-second decisions: Navigating the fine line between man and machine
by Tyler (Robotics News)
March 25, 2017, 10:49:29 PM
Most Human Like Android Built to Date
by 8pla.net (AI News )
March 25, 2017, 10:24:26 PM
Robotic science may (or may not) help us keep up with the death of bees
by Tyler (Robotics News)
March 25, 2017, 04:48:41 PM

Users Online

16 Guests, 0 Users

Most Online Today: 30. Most Online Ever: 208 (August 27, 2008, 08:24:30 AM)

Articles