Public Lab Research note


GSoC Update - AutoKnitting

by xvidun | June 20, 2014 10:59 20 Jun 10:59 | #10594 | #10594

Recent work:

It's been sometime since I've posted an update as it took some time to actually figure things out for implementing the auto-knitting feature. For the past weeks I have been refining the upload interface and beginning works on implementing the "AutoKnitting" feature for mapknitter with good progress. Although the implementation of the "Management Tool" in the upload interface from the discussions is stalled for now where I am trying to complete the core component for implenting auto-knitting. This way I can integrate this feature into the upload interface later on.

My attempt and results

So for implementing auto-knitting(using GPS data from images) I've been using cuas-ballon-mapping for reference as it contains all the necessary GPS data.

I'll use two sample images that are almost overlapping to demonstrate the results.

The map for the above reference image. I have marked the approximate positions for the sample images used.

Map-cuas-marked.png

So when I try to upload the images I get the following result.

Knit1.png

Knit2.png

Which is almost close to the actual region, I have positioned the image only based on lat/long so correction of image based on rotation and scale does not happen - I will be looking into those in the coming days.

Ideally right now I must be able to perfectly knit just by rotation and scale but for the above sample images that does not happen, which comes down to either my error in calculation or GPS data accuracy which I will investigate in the coming days. Also to verify my results it would be great if some one could share additional images(with GPS data).

When the zoom level is decreased there seem to be a marginal error in the placement, the image however overlaps the required region - looking to verify this.

Questions and next steps

In the coming days I will work on implementing the image correction based on rotation, alritude and tilt data.

Also since some work has been accomplished I'm curious about the progress on Anish's project so that we can discuss on integration.

EDIT: Forgot to link code: AutoPlace


6 Comments

Wow. Nice progress.

Reply to this comment...


This is very impressive. Thanks, Vidun.

Also to verify my results it would be great if some one could share additional images(with GPS data).

I'll ask on the grassrootsmapping list -- but Pat - do you have an image set which has GPS exif data?

I believe any iPhone image will have built in exif GPS tags. Here should be an example: http://www.brianklug.org/2011/11/a-quick-analysis-of-exif-data-from-apples-iphone-4s-camera-samples/

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


I'm asking people to post photos to this wiki page: http://publiclab.org/wiki/mapknitter-auto-placement

I also posted some images from an Android Nexus 5 which may have gps tags.

Vidun - if we get you access to a test server, can you post your code so others can try it out?

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Also, vidun - can you please commit your code to Github so we can see it? That will help with integration too. It's extremely important that you commit regularly. Every day you write code, you should commit and push something! If you want, you can push it to a custom branch called "auto-placement".

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Real cool!

Reply to this comment...


if we get you access to a test server, can you post your code so others can try it out?

That would be great, I've tried using Heroku where I had some issues using pgsql which was taking too much time to resolve.

can you please commit your code to Github so we can see it?

I've updated the note with the link, I usually push my commits atleast once in three days you can always find them at my fork in one of the working branches.

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Login to comment.