RIT Internship Reflection

Today I spent my time learning something new, Python. I had briefly started Python in the first week but this was a deep dive into the language. The very first task I had was to create a new virtual environment in the anaconda and customize it to match my mentors. Once that was done I could really explore the language. Python is organized similarly to Java, but the notation is definitely simpler. For example, if I want to take an input from a user I don't need to import an new library and can simply call input() function. I spent my time converting old Java programs I did into Python.

The eventual goal of our research was to test the effectiveness of neural networks in aerial tracking networks. In order to test this data we need to simulate the data of a town. Today I was working on creating this town from which we can get data. It is definitely a change from cars as the main challenge in this model is creating irregular yet geometrically constrained curves in 2D. It is a new change of pace.

Today is the Undergraduate Research Symposium. On a few building in the campus Undergraduates are presenting their research. I took breaks throughout the day to learn about the other research being done on campus. It was interesting to learn about the other departments. An electrical engineer was exploring new ways to use robotic arms in their lab. A student was studying the economic diversity in artificial wetlands and how easily they fall to invasive species. There was even a model of the whole campus done by architecture students. It was a real eye-opener. While I was researching aerial car tracking, people two buildings over were exploring wetland or robotics.

Today we created the first scene we were going to use. It is a representation of the tollbooth on highway 490, just to the east of the RIT campus. I started by making the seven lane tollbooth based off of dimensions of the cars and images from google maps. From there I made the road for the scene. By matching the origins and exporting the models as obj files, it became easy to align the objects in DIRSIG.

Today my mentor had a meeting with his professor about his project. To start with my mentor wanted me to create a scene of a local mall to show his professor what I had been doing. I created a strip mall and a large outlet type mall. It was a pseudo-scene to demonstrate what I had been doing to Dr. Hoffman. The meeting was the first time that I was able to learn about the convolution neural network my mentor was using. He was testing with semantic segmentation, dropout, and squeeze and excitation.

Today my focus was on buildings. I originally tried using Google SketchUp for the buildings as it is a program for architects, however the grouping ended up not working out so I remade the structures in Inventor. The general structures were basic and I focused on the roofing. In my mentor's research it is important that there are different types of roofs. We need the traditional type of roof - shingles - in addition to a terrace and glass dome. Each one of these different types of roofs would have a different hyper-spectral reading. I spent most of today making these different kinds of roofs for building as well as the building itself. It was a nice change of pace.

Today we visited the UAV Lab in the morning as a special treat. They focus on drones and aerial footage. Interestingly enough they use five different cameras on the flight. They have the basic 3 band RGB camera, as well as a 5 band and hyper-spectral camera on board. They also have infra-red and a distance tracker. One flight can give them a huge amount of data. Even with this the lab wasn't satisfied so they are configuring and new eight rotor drone with a cooled infrared camera so that they can detect the exact temperature on the ground. It was interesting.