Road sign recognition using affine moment invariant*
Choong, Yeun Liong, (2004) Road sign recognition using affine moment invariant*. Journal of ICT, 3 (2). pp. 59-76. ISSN 1675414X
Universiti Kebangsaan Malaysia. Faculty of Science and Technology. School of Mathematical Sciences
Ability to recognise road sign in due time is an essential aspect of safe driving especially at night time. Hence an auto road sign recognition system is a desirable add-on to a night vision system. This paper presents the use of affine moment invariants (AMIs) as the invariant feature vectors, and the multilayer perceptron (MLP) neural network as the pattern classifier, in developing a road sign recognition system. Six classes of road signs of different position, size and orientation, which were extracted from various near infra-red (NIR) road scenes, had been processed to validate the system. The first four simple AMIs, I1 - I4, were used for the image registration. The AMIs, which were computed from central moments, formed a feature vector that was invariant under the general affine transformation. These feature vectors were then fed into an MLP neural network for classification. The MLP used was trained with the quickprop algorithm (QA), a variation of the standard back-propagation (BPA) algorithm. Scaling and transformation of the feature vectors has reduced its dynamic range significantly towards improving the network convergence and performance. The trained and tested MLP was then validated with a set of feature vectors that it has never “seen” before. This early work has achieved a 100% successful classification rate using a limited validation set of road sign images.
|Additional Information:||(Notes: There were typos due to the scanning process in the other copy (and the pdf version of the full paper). Will be corrected soon.)|
|Keywords:||Road sign recognition, affine moment invariants (AMIs), multilayer perceptron (MLP), quickprop algorithm (QA)|
|Subjects:||Q Science, Computer Science|
Repository Staff Only: item control page