Medication Monitor - Part 1

RE: Medication Monitor - Part 1

shabaz — Thu, 05 Dec 2019 17:28:46 GMT

Hi!

This is a really good idea, to make a desk lamp more 'smart' in this way and equip it with vision. Despite not completing, you've got quite far. OpenCV is not easy to use, there is a learning curve to it, so it is not surprising that more time is needed. Great effort, and even if it doesn't get completed for medication monitoring, I think it has other uses, like seeing who is using a desk. That has real-world applications (seeing how many people are in a meeting room or how much activity there is, so that the environment can be better controlled).

By the way, is the source code for this project available? (Just curious).

RE: Medication Monitor - Part 1

vimarsh_ — Tue, 19 Nov 2019 11:21:14 GMT

Its a great idea. According to me using magnets to put it in one place is better because it is always there at the same place and you can't forget it.

Shadows and Texture on table. etc interfere a lot in image processing. OpenCV usually just treats it as a different object.

Excited to see it in action!

RE: Medication Monitor - Part 1

three-phase — Mon, 18 Nov 2019 16:38:39 GMT

Great idea for a project, hope you can get it all up and working.

Kind regards.

RE: Medication Monitor - Part 1

genebren — Mon, 18 Nov 2019 16:02:51 GMT

Interesting results. Shadows are always challenging when dealing with image processing. I have not used OpenCv before, so my suggestions are untested in your environment, but still may have some value to you process:

1) I have managed similar issues with shadows and non consistent illuminations by using an intensity correction phase early on in the image processing. I broke the image up into a series of regions and created a histogram of the intensities. Looking at the range of values that I consider background level, I determined the average background for each region. I then compared the result of adjacent regions, trying to eliminate and bias caused by non-background object. If a region had a large shift in background, I instead averaged all adjacent regions to determine the value for this region. I then corrected the image by using the delta of the average background compared to a normalize value to determine the adjustment value. Using this approach I was able to correct the image, by removing the effects of uneven illumination on the image.

2) Reject found objects that are smaller than the expected size, or are outside of the target area.

3) Zoom in on the area of interest to eliminate the errant objects being detected.

Best of luck in getting you project to work as expected!

Gene