This week I focused on finalizing what hardware we want to use. To do this, I came up with several requirements for our system. I determined that based off the average speed of a cat, our computer vision and machine learning algorithms will have approximately 1.2 seconds of cat visuals. In addition, I believe that we want to maximize the number of images we can process during this time. For example, based off my research a Raspberry Pi can compute at a rate of 1 frame per second. In most cases, we would take only one image of the cat, which is a great risk because the cat could be looking away or a light glare for just that one image. I also looked into an Odroid, which is essentially a more powerful Raspberry Pi. Even this would yield 2-3 frames per second. Again, we would be banking on receiving a stable image during these frames. Based off this research, our team decided that GPUs was the best course of action.
I focused my GPU research on Nvidia GPUs as I have experience writing parallel code in Cuda on Nvidia GPUs. Nvidia has a family of GPUs called Jetson whose application are for embedded systems. They have 256 cores. In addition, the development kit has a quad-core Arm CPU, Wifi capabilities, and many I/O ports. The Jetson TX2 was not only a solution for our image processing, but also for our system communication. In addition, I added this information along with more details to our design paper.
I also made progress with the app design, starting with a simple wireframe in Xcode:
My progress is on schedule.
In the upcoming week I will be working on finishing up the design presentation, in addition to figuring out more details about the app and the camera to Jetson communication mechanism which I will be reporting in the design paper.