My project idea initially was to transform water puddles into portals to another universe. When I went out after rains and clicked water puddles (shown below) , I realized few things:

[gview file=”https://ems.andrew.cmu.edu/2015/wp-content/uploads/2015/04/IMG_20150407_093232123.tif”]
[gview file=”https://ems.andrew.cmu.edu/2015/wp-content/uploads/2015/04/IMG_20150407_090727608.tif”]
[gview file=”https://ems.andrew.cmu.edu/2015/wp-content/uploads/2015/04/IMG_20150407_0932321231.tif”]

1. The perfect puddle(blue sky reflected clearly) does not exist in Pittsburgh. After rains, the sky tends to be cloudy. Hence , even if the reflection is clear, it is never blue. I was planning to use the distinction of color between dark, grey pavement and sky blue of reflection to recognize the shape of puddle. Looks like it is going to be extremely difficult.

2. Assuming the sky becomes magically clear and reflects blue, the puddle reflection is not going to be nice blue of sky but a blue with a tint of the pavement color (water is transperent!). Hence, this strategy does not work.

While clicking puddles I came across manholes. These are usually of same size, circular and can be found easily at many places and pavements. Surprisingly, manholes in Pittsburgh have various patterns.

I am going to detect these manholes and transform them into portals to another world when you see them through phone application.

Where I am at:
1. Golan introduced me to ImageJ http://imagej.nih.gov/ij/
I processed 3 different manhole that I have and came up with a workflow of recognizing manhole out of the image to create masks.

Manhole type 1:
Workflow:
1. Reduce the image size (400xY)
2. Convert the image into 8 bit
3. Make Binary
4. Fill Holes
5. Run Minimize filter with pixel size 7.
6. Convert to Mask

Manhole type 2:
1. Reduce the image size (400xY)
2. Convert the image into 8 bit
3. Run the Median Blur, pixel limit 5
4. Variance with 5 pixel radius

Manhole type 3
1. Reduce the image size (400xY)
2. Convert the image into 8 bit
3. Maximum
4. convert to Binary
5. Erode 5-6 times
6. Fill holes

Future Step:
Recreate detection of manholes using ofxcv in openFrameworks.

Sidenote:
I also secretly want to be transported to Japan for this project. Japanese manholes are art pieces! Look at them, they are great image targets for unity!!