Weekly Update, September 23 to September 28

Jason Huang

Accomplishments:

  • Read 3D Face Recognition on Point Cloud Data, an in-depth explanation of the facial recognition algorithm that I will use this semester
  • Modified Viola-Jones’ algorithm, which uses numpy, scipy, opencv, sci-kit to pre-process all the facial points (work-in-progress)
  • Implemented the conversion between 3D facial points to face curvature maps via matrix algebra

Upcoming Work

Within the next week, I would like to get a better grasp of the Viola-Jones algorithm that helps me pre-process all the facial points. I’ll get familiar with openCV and play around with it, and tweak the code a little more to make it compatible with other steps of the facial recognition algorithm provided by the paper. I will give myself the whole week to do this, along with improving the matrix algebra code, giving myself enough extra time to account for possible debugging.

Schedule

I am currently on schedule. There are four parts to the algorithm and I plan on tackling each section on a weekly basis. The matrix algebra covers one of the four parts, but since other parts may be harder, I would like to complete at least one part, and get started on a third part of the algorithm by the end of next week.

 

Emily Wong

I spent this week specing and purchasing parts (microcontrollers, bluetooth modules, cameras, etc.) for the front part of the door lock. These parts will be used in conjunction to get facial data and send that data to the back part of the lock for processing. There was a small speedbump with buying our camera from Intel RealSense so that was stressful, but we found a replacement (thanks Quinn!!) with basically the same specs and everything so that was relieving. Another speed bump that we encountered was that I didn’t realize that using a breakout board for USB was a bad idea to get the camera data out, since I forgot about the USB protocol data that would be intertwined with the camera data. I’ve found a few USB host shields, but we’re holding off on buying that for now to see if we can get any intelligible data using the RX/TX pins on the microcontroller instead. 

Progress is on schedule so far. This weekend I plan on designing the circuits to connect all the parts together, and next week I plan to start actually wiring things together so that we can hopefully get some camera data to Jason so that he has something to work with instead of just guesses on what the data will look like. I’m also going to call Home Depot to see if they can make us a mini door so that we aren’t bound to the lab when we want to work on our project! 🙂

 

Joe Zhao

This week I researched and brainstormed on the design for the circuit that will inductively power the front of the lock. Given some advice from Professor Kelly and Charles, instead of having to send peak power through the door at all times, we will be sending a lower amount of power, and use a capacitor on the front to handle providing additional power when the sensor is firing. With some rough math,  I have spec’d out the mosfets to use to do the switching, as well as a frequency. I haven’t bought the parts yet, because I want to check my design with Professor Kelly and get the ok before committing to parts. 

Progress is on schedule, after checking with Professor Kelly on Monday, I will order parts and begin assembly. I hope to deliver a working power delivery system that can send 2W. This is due to our camera requiring 500mA @ 5V to take a picture, and then similar power for the microcontroller sending the data. Since we want the whole process to take less than 5 seconds, we need to send at least 2W to recharge the capacitor in time to take and process another photo.

 

Whole team

After receiving feedback on our initial project proposal, we had a meeting to further flesh out our requirements. None of our original requirements changed, we just added some more to make sure that we knew exactly what we needed to get done (ex: For security, we want more false negatives than false positives [97% accuracy w/ false positives, 93% accuracy w/ false negatives based on the facial recognition paper we’re basing our face ID algo off of], since it’s safer to deny more permit). 

 

Leave a Reply

Your email address will not be published. Required fields are marked *