Next: Software design Up: Conclusion Previous: Conclusion Contents Index
The recognition system was implemented in Perl, C, Matlab and shell-scripts. The first implementation of SIFT was done in Perl and using Image::Magick library for the basic image manipulaiton such as loading, saving and color channels operations. The SIFT computation took approx. 10 seconds for a geometrically normalized image (256x92 pixels). By rewritting the SIFT module into C with support of libpng for basic image operations we gain a 70-150x speedup.
Other improvements to the classification rate can be achieved by using a different classification methods such as Support Vector Machines or AdaBoost. The proposed method would need to be evaluated with more vehicle sample images.