Shaposhnikov D.G.1, Podladchikova L.N.1, Shevtsova N.A.1, Golovan A.V.1, Gao X.W.2
1A.B. Kogan Research Institute for Neurocybernetics,
Rostov State University, Rostov-on-Don, Russia
2School of Computing Science, Middlesex University
London, United Kingdom
A biologically plausible algorithms and model for traffic sign recognition invariantly with respect to variable viewing conditions are presented. They simulate several key mechanisms of biological vision, such as space-variant representation of information (reduction in resolution from the fovea to the retinal periphery), orientation selectivity of the cortical neuron responses, search for the most informative image regions, and context encoding of information. Traffic sign processing procedures include colour segmentation, classification according to sign contour colour and shape, and finding the sign centre for single positioning of a space-variant sensor window. It has been revealed that recognition rate is relatively high for signs under artificial transformations that reproduce possible sign disturbances in real road conditions (up to 50% for noise level, 50 meters for distances to signs, and 5° for perspective disturbances). As compared to others, recognition rate for red triangular signs sharply decreases at these distortion levels. While processing the British real world traffic signs (n=98) obtained under various environmental conditions, the recognition rate is equal to 0.95.