DiffCam – demos related to motion detection with JavaScript

I have been working on a binary image diffing algorithm for a project called Narcis (https://github.com/deckar01/narcis).

The idea is to provide visual regression testing for integration tests. It's not a new concept, but the weakest part of the solutions I have seen are really ugly diffs. I am also playing around with the idea of streamlining a manual review UX instead of the standard change threshold concept.

I have mainly focused on assumptions that can be made about screenshots of a webpage which allows more fine-tuned diffing behavior.

Demos:

https://narcis.us/test/google/master/home/000000/diff/000001

https://narcis.us/test/google/master/home/000000/diff/000006

https://narcis.us/test/google/master/home/000000/diff/000009

Excellent demos. One suggestion for the creator/maintainer of the demo pages: It would be nice if the imageDiff demo had two pre-made images to do a diff on. I spent a few minutes taking a shot with my webcam and then another with my had in the scene just to see how the static diff worked. Pre-loading a shot of a background and one with something added to the image would be a nice start to the demo instead of making the user drag/drop their own images.

As a totally unrelated note, if this interests you, you might also get a kick out of coherent change detection (CCD) which works on synthetic aperture radar (SAR) images: http://defense-update.com/features/du-2-05/sensor-5.htm. CCD has been used by military intelligence for a while now, and although a lot of the cool stuff is classified, the basics of the technique are unclassified and kind of cool.

I spent Memorial Day weekend working on Python/OpenCV traffic counter that uses similar motion detection techniques. I tried to explain how it works in the readme on the GitHub repository:

https://github.com/iandees/speedtrack

For anybody doing similar things in javascript, you might look into jsfeat: https://inspirit.github.io/jsfeat

It provides some basic building blocks for interesting computer vision in js.