ModelFly Project
The Reichardt Detector
Zoom Image
Reichardt Detector. Click for more information.
The direction of motion is not explicitly represented in the activity level of the photoreceptors. In response to a moving grating, the photoreceptor signal is just modulated according to the number of stripes passing by per second, no matter in what direction the pattern is moving. Only when at least two photoreceptor signals, displaced along the orientation of image motion, are considered, the direction of motion can be derived by an external observer based on the shift of the signals relative to each other. Thus, motion vision represents a simple but well-defined example of neural computation. A principle mechanism of how the direction of motion can be computed from the signals of neighboring photoreceptors was proposed by Reichardt and Hassenstein in the 1950ies. Their model, meanwhile just called 'Reichardt detector' in the literature, incorporates two essential operations: asymmetrical temporal filtering (LP, HP) and a nonlinear interaction stage (M) where the low-pass filtered signal from one image location is multiplied with the high-pass filtered signal from the neighboring image location. At each image location one needs at least two such units with orthogonal orientations.
Based on abundant experimental evidence (e.g. Single S, Borst A, Science 281, 1848-1850, 1998), we assume that such operations are indeed taking place in the fly visual system. The neurons performing these operations, however, have not yet been identified.
In our simulations, we pass visual images through a 2D array of such detectors and feed the output signals of the two subunits for each of the two orthogonal directions as conductance values for excitatory and inihibitory synapses onto dendrites of our models of the lobula plate tangential cells (see Lobula plate chapter).
