This week, I prepared for the interim demo by ironing out the final kinks in the facial recognition software. Most notably, there was a bug that caused recognition to occur when almost all of the frames currently saved still included the face of the previous person (basically, it was trying to do recognition while not updating the list of recent embeddings, causing it to still guess the previous person). I was able to solve this by making sure to update the recent embeddings I was keeping track of so that they align with the recent frames (something I’m also keeping track of).
When it comes to facial recognition I am on track, but the web app will become my primary focus to fully implement before carnival.
As I stated above, next week will be solely focused on getting the web app fully up and running before Carnival.
For verification and validation, I currently plan on testing facial recognition in two ways, with a static image database and real-time facial recognition (as each has their requirements that I must make sure they are fulfilling). Firstly, testing is needed to make sure the recognition is accurate enough based on the design requirement (which is currently 95% accuracy). To do this, I plan on using a well-known face database and gathering a collection of 20 faces (each face having some amount of pictures of them) and splitting the dataset into training and testing sets. Then I will use the recognition system to first train on the training set and then introduce the testing set to see how well it can recognize the new faces.
As for real-time facial recognition, our design requirements state that the system must be able to detect faces within 0.5 meters, within 5 seconds (while still keeping 95% accuracy). I plan on testing this by marking different distances away from our camera (distances both within and outside 0.5 meters), and then timing how long it takes to detect my face. To still make sure we are meeting the 95% accuracy requirement, I plan on repeating this test on at least 20 different faces.