Guide Sign Analysis of Traffic Sign Data-Set Using Supervised Spiking Neuron Technique

Abstract

In this paper, 20 guided traffic signs mostly displayed around Malacca area were selected as project databased. Early hypothesis was made as the error for each usable image will increased as more interference introduced to the original image used. Three types of conditions which are hidden region, image brightness and image rotation were selected as an experiment to analyze the performance of each sign used. Each condition will perform a specific error to generate their mean value and in the same, image recognition will take place in the matchup process. By focusing on the result, it produces hidden region critically ascending mean error value at 62.5% = 0.07 and has average value at others points. For image brightness effect, it shows a higher mean error value collected at less brightness points and non-stable pattern at 10% to 60% brightness. As for rotation upshot, the values show a critically ascending for error value at 22.5% and slightly increase at 2% to 5% rotation point. For the recognition process, at 6.25% hidden region, almost 70% of images are correctly matched to its own classes while at 62.5% hidden region only 40% of images are correctly matched to its own classes and leaving 2 images to outperform. For -40% brightness, 45% of images are correctly matched to its own classes while at 60% brightness 65% of images are correctly matched to its own classes and leaving 1 image to outperform. Lastly, at 2.5 degree rotation, 85% of images are correctly matched to its own classes while at 25° rotation, 45% of images are correctly matched to its own classes and leaving 2 images to outperform. Finally, the error forms will affect the final output response of the detected traffic signs used.