Updates:
Facial Recognition: (Ryan, Kurtis, Brad)
We still have yet to resolve the issues with false positives. We have tried playing with the threshold values, as well as using different face recognition algorithms supplied by OpenCV, but still occasionally get false positives. We have also tried taking more pictures for the data set when setting up the training, but this also did not solve our issue. By process of elimination, we think it has to do with the negative images used in the training data set. Negative photos are used so that the algorithm knows what a bad match looks like. We have a hunch that with a better negative dataset, we will eliminate false positives in facial recognition.
This week, we were also able to further develop our training module. A simple button, in lieu of the Hover sensor, was implemented so that users can control when to start training after registering their account on the web app. A new picture-taking protocol was also implemented where the mirror takes a quick 20 initial pictures, and then post-processes these in an attempt to find 10 usable pictures. This allows the user to just stand in front of the mirror for a short period of time in order to train their face into the mirror. The process will repeat until the mirror is able to save 10 usable pictures.
Mobile Application: Updates (Brandon)
The physical button that users were using to begin the face recognition training is now incorporated into the Mobile App as well. We were having difficulty thinking of a way to conveniently place the button onto the Smart Mirror Frame without a breadboard during our last meeting. In order to resolve this issue, we decided to incorporate it into the Mobile App as seen below.
Gesture Control (Ryan, Kurtis, Brad)
The button has been set up and integrated into the face recognition training protocol. Now, when a user registers using the mobile app, they are prompted on the mirror to start the training by pressing the button. Since we were not able to get our other gesture sensor to work, this was our workaround to have a basic UI for the mirror.
Smart Mirror ON/OFF Protocol (Brad)
No Updates
Smart Mirror Frame (Ryan, Kurtis, Brandon, Brad)
This last week we tried applying the mirror film to the acrylic panel. After two attempts of applying the film, we were able to get decent results. There was some bubbling but it was not very noticable. However, since we didn’t have a squeegee we used a piece of plastic to try and smooth out the film, which resulted in some scratches. We have plenty of film left so we are going to do another run to get the best results possible.
Next Steps:
Facial Recognition: (Ryan, Kurtis, Brad)
Since lightning and room conditions may impact facial recognition, we are going to do some test runs for facial recognition in the final demo room. To populate our negative training set we will need to have a few people not in our classroom volunteer to be the negative photos. Once we have gotten a few volunteers to gather these negatives, we can try running through our demo with as many people as possible to test for issues. These next couple weeks will consist of lots of fine tuning and testing to make sure our facial recognition works properly.
Mobile Application: (Brandon)
The last part of the Mobile App is to save the customizations of each profile and to pull up the settings when the user is recognized by the Facial Recognition. Currently, the settings that users are able to set are universally shared amongst all users. In order to address the more crucial false-positive issue, we may axe this feature of the Smart Mirror.
Gesture Control: (Ryan, Kurtis, Brad)
We may look into adding more gestures using the button, such as a long press to clear the screen. This is not a priority since it is not really part of our final demo, but we will still look into it if we have time.
Smart Mirror ON/OFF Protocol (Brad)
The proximity sensor needs to be integrated back into the mirror using a breadboard so that there are enough connections for all peripherals. The main code has already been written so it is simply a matter of connecting it back up and tweaking the values to our liking.
Smart Mirror Frame (Ryan, Kurtis, Brandon, Brad)
We are currently still taking measurements and planning the layout of the mirror frame. Since our budget is low, we want to make sure that we get the dimensions right the first time so we do not waste materials.