Arduino/embedded vision and camera brain dump

arduino_vision.markdown

Broad considerations:

• lens -- most sensors have 30 degree+ lens installed. For narrow FOV, either read just central pixels or use a lens.
• sensor -- fast readout, low power, few pixels so the arduino can handle the data
• otf data preparation -- need to do flat-fielding, hot/cold pixel rejection on the microcontroller
• algorithm -- threshold or something simple to find white spots on black background
• data stream -- send just locations of first n detected white pixels. Make sure it works with screen edges, etc.

Pre-positioned point lights.

• Many small lights
  • small FOV sensor (pencil beam)
  • detect local asterism, determine which part of the screen/field we're looking at
• Few large lights
  • large FOV sensor (pinhole)
  • detect global asterism, determine where camera center is pointed

Centeye embedded vision site

Locating lights using ardueye

Sensor options:

• $10+ Sparkfun options mostly vga-ish, some with jpeg encoders

• $40 Adafruit TTL camera 30fps vga video, with image capture on demand

• $100 CMUCam VGA with on-board image processing (full-size Arduino shield)

• $50 CentEye Stonyman 112x112, on-chip binning, addressable readout (single, line, region), asynchronous, logarithmic pixels, Prototyping shield