Now I work at associative video memory. The method still in developing (now it version 0.5)
but it gives good results already today.
I am dealing with research of computer vision in parallel with my main job
at "Impulse" more than three years (it is my hobby).
About me: http://edv-detail.narod.ru/EDV_resume.html
In the beginning my achievements were insignificant and little part of ideas has worked properly.
But I did not surrender. I generated big quantity of hypotheses and then test it.
The most ideas did not work indeed but those that worked were similar to particles of gold
in huge quantity of dross. My associative video memory method is working indeed.
============================- Common information -==========================
Algorithm AVM uses a principle of multilevel decomposition of recognition matrices,
it is steady against noise of the camera and well scaled, simply and quickly
for training, also it shows acceptable quick-action on a greater image resolution
of entrance video (960x720 and more). The algorithm works with grayscale images.
The detailed information about AVM algorithm can be looked here:
AVM SDK v0.5 with examples of using and tests for comparison
of characteristics of the previous and new versions:
Demonstration video how to train AVM:
AVM demo with the user interface (GUI), installation for Windows:
Connect the web-camera and start AVM demo after installation of "Recognition.exe".
After starting the program will inform that there is not stored previously data
of training AVM and then will propose to establish the key size of the image
for creation of new copy AVM. Further train AVM using as an example Face_training_demo.avi.
========================- Robot's navigation -=========================
I also want to introduce my first experience in robot's navigation powered by AVM.
Briefly, the navigation algorithm do attempts to align position of a tower
and the body of robot on the center of the first recognized object in the list
of tracking and if the object is far will come nearer and if it is too close it
will be rolled away back.
See video below:
I have made changes in algorithm of the robot's control (Navigator.cpp)
also I have used low resolution of entrance images 320x240 pixels.
And it gave good result (see "Follow me"):
Source code of "Navigator" program:
During experiments I noted significant reserve of quick-action
when AVM v0.5 is working with resolution 320x240.
And it give me good conditions for starting of new experiments
in navigation field (algorithm "Location tree"), of course I also
will try to make AVM faster with resolution 640x480 in the future.
Also I finished some experiments with "Natural beacons":
Source code of demo (project "Beacons test"):
I hope that this information can help you in questions:
How we can use AVM algorithm?
If somebody can read in Russian then welcome: (Russian thread: "Autonomous robot's navigation").