We’ve had an exciting week of taking all of our brainstorming and researched methods to the streams. We started with a trip to Menard’s to find materials we thought would be durable in the stream environment, portable from the classroom to the stream for repeat trials, and accurate for horizontal and vertical lines to be used in Hugin photogrammetry software.
We decided to construct an 8’x8′ square to be placed near the surface of the stream. Upon arriving at our Duck Creek site, we were surprised to see how much the water level had risen and spread since the first time we visited almost two weeks ago. The high waters eliminated most of the sandbar banks that we had hoped to use for anchoring our square and placing accuracy testing objects. We managed to anchor the square at the top of the banks but objects for the accuracy tests could not be placed at a point that they could be seen in each picture.
We plan to use the square to perspective correct the picture and to determine an inches/pixel ratio after perspective correction. The control points we need to add will be used to compare pixel to pixel distances to measured distances as an indication of the accuracy of the perspective correction. Although we couldn’t obtain all the data we had hoped, we learned a lot about our procedure in terms of what worked and what did not work and were able to use the pictures in Hugin to get some perspective correcting practice.
Because the water level was too high, we decided to take our plans back to the fake stream to get a better understanding of what information we need in the various photo software and to determine the best way to get that information. We began by using construction paper and sharpies to create various designs of ground control points and squares for perspective correction. Once the different designs were completed we took pictures from different angles to determine the design that could be seen best from increasing distances. We voted on designs and now have 12 ground control points and 3 perspective correcting squares set around/in our “stream.”
With hopes of the water level lowering in the near future, we went ahead and made various control points and tested the different designs in a sandy area near the real stream to see which would be the most visible in pictures.
On the programming side of things for our project we have been working towards making a graphical user interface for a MATLAB program which does perspective correction by forcing pixels at the corners of our square to turn into a parallelogram on a plane in three dimensions. The Dr. Stonedahls worked through the math and Susa turned this into a MATLAB program. In order to better make use of their program a graphical side was written in Java by Kyle. This program takes into consideration the: Filename of the .jpg or .png image, the (X,Y) locations of the corners of the square A-D, the horizontal field of view of the image, and the square side length. Below you can see an image of what the graphical side looks like currently.
As you can see the program allows to browse through the computer to find the image in question and then the user inputs the (X,Y) values of the pixel coordinates which then all get inputted into the MATLAB program. Additionally this program is also capable of scanning through an image and looking for specifically colored pixels. This is important because when outlining our fake indoor stream we made the corners of the square yellow, the control points red, and the outline of the tape stream magenta. See below the tape stream and then the outline with color next to it.
These above pictures when combined with the programs are the first step to actually getting an outline of the stream which is entered into the computer. We still need to get our camera lenses calibrated and are hoping that this will improve our error estimates, which are OK, but not great on our test stream.
One of other programs we are trying out is called Photosynth. It takes pictures and lays them out according to their control points. So if one point is similar to another in a photograph it links those two photos together, and thus builds a point cloud out of all of the points matched up in the photographs. This is what is displayed below, a point cloud out of all 179 photos of the stream.The program doesn’t seem to find control points within the stream, so we get a pretty nice outline to show up. You can very clearly see the edge of the bridge at the bottom of the picture.
Lastly, we have continued to search for and read articles that have a similar project goal as us to explore alternative methods and get some more ideas about the best way to obtain, analyze and process data.
Thanks for reading! If you want to follow our project more closely, here is the link to our group’s blog.