TR-H-0226 :1997.7.17

Martin DAVID, Frans VERSTRATEN and Nicolas SCHWEIGHOFER

Model of Motion Detector

Abstract:The purpose of this model is to reproduce human perception regarding sharpening of perception, integration of similar direction transparent motion, independent perception of distinct transparent motion and motion after effect.

The front-end of the system is a vector of 24 units, corresponding to 24 directional selective cells (every 15°). This layer correspond to motion detector cells.

The output of the system is another vector of 24 units corresponding to higher cognitive level cells, namely perception cells.

The implementation of such system with a neural network needs a lot of tuning. Therefore, the program, written in C++ to allow great execution speed, is called by a MATLAB script. MATLAB handle the display and every parameter can be changed easily.

For an explanation of the model, its equations and the link with psycho-physics results, please refer to Alexander Grunewald's papers. For further detail on the mathematical model and the link with biology, please refer to Grossberg's paper.