1. Assistive Technology for Patients with Medical Face Blindness
Project 10
Timothy Chia - tchia2
Jeffrey Luan - jrluan2

2. Introduction
Prosopagnosia - A neurological condition which affects the ability to recognize familiar faces
As a result, social interactions are adversely affected
Workplace
Family and Friends
Safety

3. Objective
Facilitate social interaction for individuals suffering from prosopagnosia
Utilize existing smartphone technology as effectively as possible, while remaining unobtrusive

4. High Level Modules Camera Processing Wristband
Capture and transmit images to the processing module
Processing
Identify faces in photos
User interface
Communication hub for other modules
Wristband
Display names and vibrate in response to information from processing module
Button for user to request a recognition cycle

5. Design

6. Software Cycle

7. Project Build

8. Additional Design Considerations
Low cost
Familiar user interface
Portability
Battery Life
Development Time
Iterations

9. Modularity Hardware Software Breadboard Arduino
Computer instead of Android phone
Battery Circuit Independence
Software
Testing/Training of API on computer
Android app functional independence from hardware modules

10. Stage Targets Recognition and database capabilities on computer
Hardware component functionality achieved on breadboard
Android app functional without hardware
Android app communicating with hardware
Android app fully functional with hardware
Android app fully functional with PCB hardware

12. Recognition Design Kairos API Inputs, Outputs, Interface, Capabilities
Threshold control
Training
Calibration:
Priority on minimization of false positives

13. Facial Recognition Requirements
False Positive < 5%
Training Set: 7 individuals, 15 photos
Testing Set: 13 individuals, 1 photo
Constraints:
Distance = 1.25m
Angle = Direct
Lighting Conditions = Indoor, all lights on.

14. Receiver Operating Characteristic Curve

15. Challenges

16. Bluetooth Receiver Issue
Android Buffer Issue
For unknown reasons, bluetooth inputstream consistently crashes after receiving about 15,000 bytes, refusing to accept more bytes from the HC - 05.
Diagnosis
Attempt transmission of various size arrays from HC-05, also varying the android side handling. Attempted several varieties of continuous reading, buffered reading, blocking reading, parallel thread reading etc.
Significance
320 * 240 = 76,800 > 10,000
Conclusion
Must settle for 10 kB chunks

18. Bluetooth Transmitter Issue
HC - 05 Speed Issue
Serial input below datasheet spec. Maximum of 115,200 baud instead of 1,000,000 baud.
Diagnosis
AT command rejected.
Significance
Images received by computer from camera become garbage at this speed.
Interrupt Handler Interference
Images became garbage after implementing asynchronous command reception from phone module.
Can selectively disable interrupts during pixel capture.

19. Bluetooth Solution
Buffer Issue: No memory in PCB layout. Downscaling by 1/16 is 4800 bytes per frame < 10,000, but only have ~1000 bytes of memory in microcontroller.
80x60 below API recommended resolution.
Speed Issue: Cannot transmit 4800 bytes synchronously either.
No time for hardware solution.
SPI SRAM ordered, circuit revision drafted in case, code drafted for TIMER1 and SPI pin shuffling.

20. Pixel Timing Original driver transmitted each byte synchronously.
(blocking on UCSR). Low speed means transmission desyncs from PCLK.

21. Row Timing Use PCLK to capture ¼ pixels in this row.
Resynchronize. For our speed, need 2 rising edges.
Asynchronously transmit during next 3 rows.

22. Example Camera Issues:
Pixel synch lost due to low speed serial transmission
Correct # of HREF blocking waits
Row synchronization incorrect: (synch, asynch without HREF, asynch with wrong # HrEF

23. Hardware Challenges PCB Voltages below specification
Diagnosis/Potential Causes
Multimeter on regulator circuits without load, and stable 5V input.
Voltage regulator damage (possibly due to static discharge or other human error)
Conclusion
No time to replace due to small form factor. Bypass our power circuitry entirely.
Camera circuit not returning data
Potential Causes
Mismatch between PCB design and breadboard prototype
Initially, used breadboard to reorder pins and add pullup resistors
Camera eventually went completely dead, even when returned to original circuit; unknown cause, but most likely damage inflicted onto camera during testing

24. Hardware Challenges (continued)
Oversights in PCB design
Diagnosis
After receiving PCBs, the following errors were noticed
Certain traces to microcontroller were wrongly placed
USB port placed out of place
Conclusion
While not an fatal error, took time to troubleshoot
Would have corrected provided more time/board submissions

25. Conclusions Communication between all modules
Image capture and recognition
Phone App for database management
State machine and command transmission for modules
3000 lines of code ready for next iteration of hardware modules

26. Future Work Size and Enclosure Image Quality Battery Circuit
Delays encountered in PCB development
Would have selected smaller components and optimized our PCB design to be smaller (for realistic portable usage)
Image Quality
Because data errors occurred at 320x240, we had to downscale images. Severe loss of accuracy.
Battery Circuit
Voltage regulator issues prevented battery use
Troubleshooting required
Replacing regulator(s)
Redesigning circuit