saltzmanj commented 8 years ago

Monday, September 12th

LabVIEW has camera interfaces: image datatype
- include large palette of tools, but not all can go on myRIO
- can be emulated on PC, if the elements are available also on myRIO
Machine Learning Toolkit for LabVIEW available on NI Website
Python window also available
Image Recognition
- Print out music at high contrast
- Avoid glare: backlighting
- usually works well for binary pixel classification
- Tools available to map shapes to each other
- Vision library is pretty huge
- "Blob analysis": very potent tool
- can define custom threshold
- Need to think out fiduciary marks
- Investigate lensing of cameras
- Find cameras that work well with myRIO
- Telecentric lens: probably too expensive
- First functions for processing: find a region of interest
- Next: resizing
Gantt Chart
- Need to allow time for synthesis and testing: make that a new task(s)
- Keep in mind this is a living document
- Add sub-tasks, such as for PCBs
- Include testing times
PCB/Hardware
- Come up with a testing plan, order a few before all 44+
- Include this in the Gantt
- Think about power and signal connections, feedback
- Needs a test sequence
After lab reveal: test plans

saltzmanj commented 8 years ago

Monday, September 19th

share Gantt file on Collab
machine learning library in LabVIEW which may be helpful
- Python or Mathscript also, but will be slower than native LabVIEW
Overall, proposal looks great

Where to go from here

NN Algorithms

Backpropogation may converge to local minima
- is a "global learning" algorithm
Other algorithms can do better
"Local learning" algorithms
- Each input only affects a certain part of memory
- Usually executes faster
- Take a look at LVQ, CMAC

Image Processing Tips

Try doing binary thresholding: no greyscale
If we can get suitable features from image processing toolkit, can try using those as features
There is a VI which will read in image files to native LabVIEW image format

saltzmanj commented 8 years ago

Optimal minimization strategy
- Look carefully at myRIO vision libraries, SVMs
- binary images
Image processing work
1. Find staff locations
2. Look for vertical lines representing the boundary of each measure
3. Check to ensure sum of notes is consistent for time signature
Starting out with music notes
1. Get images of music with cell phone, in-focus, well lit
2. Simulate camera acquisition
3. Explore high-level functions in image processing library
4. Experiment: what image processing functions work
May need to set limit for ledger lines
Can use transceiver for interfacing differential signaling with LED driver
- RS/EIA-422, RS/EIA-485: differential signaling standards
Driver board
- Include regulator: 5V/3.3V
- Use a driver board for each octave
- On last board: add shunt jumper so data can shift out to myRIO
- System can detect number of octaves
- Use through-hole LEDs: flexible cable to jumper headers to connect LEDs to driver boards
Check out laser harp
Maybe: use 24-pin ribbon cable to connect octave boards
Think about note feedback
RGB LEDs probably have 3 wires
High efficiency LEDs: probably only one color, but only uses ~2 mA
Don't connect power supplies in parallel!

MrBitSage commented 8 years ago

Powell has emailed Kira about NI Licenses
- M80X85770 Not valid
Consider 20 pin ribbon cable and make half of them ground
Investigate adequate current supply from 9V to mitigate wasted wattage / heat
Look into switching regulators instead of linear regulators (3 terminal drop in replacements for linear)
- Also, use regulators on each board instead of just the first one
Get + and - on opposite pins so that they're adjacent and have 2 grounds between each set of +-s
PCB Fabrication turnaround is about a week
No glare problem so don't need backlights
Use IMAQ instead of Vision Assistant because Vision Assistant may not work on myRIO -Deliverables for next week
- Have fiducial recognition and rotational alignment done using IMAQ
- Gerber files for PCB
- Implemented OPTICS Clustering