Development of a fuzzy neural network color image vehicular detection. (FNNCIVD) system

Abstract

This paper develops the fuzzy neural network color image vehicular detection (FNNCIVD) system to detect multiple-lane traffic flows. A pseudo line detector with fourteen detection points is placed on the monitor to detect the two-lane traffic images. On each detection point, the differencing or R, G and B pixel values between the background image and instantaneous image are inputted in every one-tenth second into a four-layer fuzzy neural network trained by the backpropagation algorithm. Traffic scenes in the daytime and nighttime are both experimented. The experiment results show that the success rates for traffic counting in different lighting conditions can be as high as 90%, in the mean time, the success rates for vehicle classification can reach 100%.