Accuracy of Visual Velocity Estimation by Reichardt Correlators

with Simon Laughlin and David O'Carroll

In my master's thesis work at the University of Cambridge, I explored visual motion detection systems computationally and experimentally.  While we wish to understand general principles of  biological motion detection, flies serve as convenient model organisms primarily because the ease of conducting certain behavioral and neurophysiological experiments has led to the collection of a wealth of data on their motion detection system.

A great deal of experimental evidence supports the Reichardt correlator, a model formulated over 40 years ago, as a mechanism for biological motion detection.  As Reichardt pointed out in his original work, however, the correlator does not signal true image velocity; correlator output depends heavily on the spatial characteristics of the visual image as well as on its motion.  My master's thesis attempts to resolve how a system which estimates velocity poorly in laboratory experiments can provide useful information in the natural world.

In particular,  I examined the accuracy with which realistic Reichardt correlators can provide velocity estimates in an organism's natural visual environment.  I found that the predictable statistics of natural images described in the last decade imply a consistent correspondence between mean correlator response and velocity, allowing the otherwise ambiguous Reichardt correlator to act as a practical velocity estimator.  My analysis and simulations also suggest that processes commonly found in visual systems, such as prefiltering, response compression, integration and adaptation, improve the reliability of velocity estimation and expand the range of velocities coded. 

David O'Carroll and I performed a set of experimental recordings which confirmed our predictions of correlator response to broad-band images.

Related Publications

R.O. Dror, D.C. O'Carroll, and S.B. Laughlin. Accuracy of velocity estimation by Reichardt correlators. To appear in Journal of the Optical Society of America A.
[Abstract]    [Full text, PDF]

R.O. Dror, D.C. O'Carroll, and S.B. Laughlin. The role of natural image statistics in biological motion estimation.  IEEE International Workshop on Biologically Motivated Computer Vision.  Springer Lecture Notes in Computer Science 1811:492-501, 2000. 
[Abstract]    [Full text, PDF]

R.O. Dror. Accuracy of visual velocity estimation by Reichardt correlators. Master's thesis, University of Cambridge, 1998.
[Abstract]    [Full text, PDF]

D.C. O'Carroll and R.O. Dror. Velocity tuning of hoverfly HS cells in response to broad-band images. In preparation.