Linaro + opencv + exercise

Exercises with Ubuntu-Linaro

1º Load an image

2º Contours

	libjpeg	libjpeg-turbo	Linaro
Load	199	63	126
2Gray	61	45	57
Trheshold	12	12	12
Create	32	11	10
find	53	31	71
Draw	47	47	87
Total	404	209	363

 3º Search one pattern

	libjpeg-turbo	Linaro
Load Source	11	6
Load Pattern	11	6
Search	488	487
Total	510	499

4º Haar-Features

The time is quite similar

Conclusion 

The ubuntu-linaro is a little bit more fast, but it'snt a big difference.

I prefer to use raspbian cause it has a big community, and the information it's easy to find, and may be it has more things that I need, but I feel comfortable  in raspbian than in ubuntu-linaro.

But I know that I'm not taking all the power of the cubieboard, I've in mind to improve the compilation using cross compilation, and take advantage of the NEON acceleration, but I'll do latter when I've more knowledge 

It's any one there ?

Well there are a couple of possibilities to check if there are any object, person

But one of the simplest is to use Haar-like features.

To work with Haar features (see the example FaceDetect.cpp in your directory OpenCV-2.4.3/samples/c/facedetect.cpp)
But basically works with haar files which store an abstract of the information about what is a face (any object) and what is not a face (a object)

Depending of the necessity and the power of our machine, we should different haar libraries. Even we can build special libraries for our purposes. (but my i3 with 4 GB ram took 3 days to make one haar library about cars)

I used this three
  

"./haarcascades/haarcascade_frontalface_alt_tree.xml" (3.5 MB) (over 500ms)
"./haarcascades/haarcascade_frontalface_alt2.xml" (0.8 MB) (over 300 ms)
"./haarcascades/haarcascade_eye.xml" (0.4 MB) (over 200 ms)


Also depends the size of the image, I used the less quality (160x120) for the first ideas will works

The code is based in the sample, so you'll find the code there, any doubt ask me.

With this exercise I finish the first part of the project based on computer vision.
We can do a lot of things more, blobs, our haar libraries, structural analyzes, movement studies, etc... but all this exercise are outside of this scope.