Joel Osei
This week for capstone we focused on our design and completing both the presentation and the document. This required a lot of talking and figuring out the fine details on how things would work for the G.Lock. Early on in the week, we focused on the presentation, refining it and making sure that Omar was ready to present. This meant that we had to do a lot of connecting the dots and making sure that we actually had things detailed out. After the presentation, we finally got back to the building, and were able to order a very important part for our project, the door lock. We are using an electric door strike which will require a simple circuit to unlock and lock. This means this upcoming week when it finally comes in we will be focusing on putting that circuit together and making sure we are able to comfortably unlock and lock. After that, we will be able to connect it to the raspberry pi and control the lock digitally.
On the other side of our project, we have to focus on actually getting the ball rolling for a lot of different parts of the project. This means we need to finalize our neural network research and begin implementing and testing, while also beginning to create the react native mobile application. Our plan is to independently develop each of these three parts and then from there test and combine everything into a final masterpiece. The biggest risk of our project at the moment I believe will be making sure our machine learning model is accurate enough. Although we have a plan that we believe will work, we understand that it will take a lot of training, testing, and fine tuning, of our model in order to successfully complete this project. We have also researched multiple methods that would work as well.
Omar Alhait
This week, I was focused on compiling the correct training set to get our algorithm off the ground. I also did a good amount of investigation for our main facial verification algorithm; we now have an accurate specification of our algorithm. It will be closely related to Fisherfaces, but a key difference is that we need to apply a layer that attempts to normalize faces for light and dark conditions. This will allow us to normalize photos with a variety of different scenes, and to maintain accurate facial profiles through those conditions.
For our training set data, I spent a lot of time taking pictures of the Raspberry Pi IR and regular cameras. I tried to take photos of different combinations of people wearing different things in order to make the span of the training set as wide as possible. We now have photos of scenarios with one person (for all our house inhabitants), different random combinations of two people, including non-inhabitants, and a few with multiple people. The next step is to start writing our actual computer vision algorithm and to connect it with our training set data to get as accurate a model as possible.
Chinedu Ojukwu
This week we finalized our design and finalized each of the subsystems that will be implemented in our project. I worked on mapping the data flow between components within the RasPi software. This includes the notification generation, the out of box streaming library, and the hardware control using GPIO and USB connections. This also includes how the system will interact with the ML testing suite that Omar will implement.
We are currently a little behind schedule. We planned to have a working first prototype of the circuit and RasPi server functionality before spring break. As of now, it appears we will have this iteration done by after spring break. To make sure we catch up to our planned schedule, we will work a good amount during spring break.
RasPi Software:
For deliverables, I plan to set up the Raspberry Pi’ server functionality. This will entail accepting requests from a client and sending data back and forth from the client and server. I also plan to implement the sending of video and image data from the RasPi server to a client. This will allow us to quantify the speeds at which data can be sent from our system to the clients.
Electric Strike Circuit:
We plan to set up the electric strike circuit that will consist of resistors, diodes, transistors and the electric strike. An electric signal will be sent from the RasPi using the GPIO pin. The additional circuit elements will be used to control power and ensure that we limit the current being send drawn by the strike.