Raspdrones

1
Raspdrones

• Using pattern recognition to make
• homes a little safer

2
What is pattern recognition?

• Extract features from data
• Use features in equations
• Equations give likelyhood of recognition

3
The problem

• Some people are not welcome in my home
• Politicians, Tax people, bailiff, salesmen
• I want to keep them out
• But how?

4
Solution

• Use pattern recognition
• Detect and recognize their face
• Interface with some hardware
• Finally a great use for the Raspberry
• We get to write some code
• And even get to solder stuff!

5
Inspired by the goalkeeper
6
Shopping list

• One raspberry
• One usb-interface board (KA8055)
• Projectile weapons
• Servo-motors
• Webcam
• Various connectors and electronic components

7
Prepping the raspberry

• Opencv
• And required libraries
• Compilers
• Waiting for the installation to finish...
• Add powered usb-hub
• And wait some more

8
Some code

• OpenCV
• Haar-features to detect faces
• Custom code to recognise the faces

9
Training the recognizer

• It watches online videos of course
• And classics from my own collection

10
Extracting the faces

• Haar-cascades
• Getting thousands of images
• Storing them
• (Manual) sorting them

11
Normalize the faces

• Detect eyes, nose, mouth, facial edges
• Rotate the face to a standard(normalization)

12
Extract meaningful data

• Measuring distances
• That's it!
• Well, actually we have to store it somewhere
• Lots of generic code

13
Recognizing

• We do the same steps as before
• But now we try to match up the data
• If there's a match, we have recognized someone
• Almost never exact match

• Sadly not enough time
• Setup is there though

14
(De-)Soldering

• Also not enough time

15
Connecting it all

• Just attach it to the usb-ports
• Run the code
• And get one frame per 10s
• Lowering video quality should improve this

16
Testing it

• Testing on a normal pc/laptop

17
Problems...

• Avoiding friendly fire
• Making it cat safe
• Make it run faster
• Better recognition
• Tons of libraries

• Friend or Foe?

18

• Presentation code available next week
• www.vankoert.org/talks/20130331
• awesome cat video
• http//www.youtube.com/watch?vPXRX47L_3yE
• OpenCV
• http//opencv.org/