An evolutionary visual landmark recognition system

A vision-based landmark recognition system by using the evolutionary principle for robot navigation tasks is implemented in this study. The research is aimed at using the GA to do pattern matching. The basic idea is to use genetic algorithms to find the best matching between nodes of the two patterns. The evaluation function can be defined in terms of total differences in magnitudes of nodes between the desired pattern and the real pattern. A search method based on genetic algorithms for pattern recognition in digital images is implemented as the vision layer for a behavior based mobile robot. The vision layer can recognize artificial landmarks by searching all the pro-defined patterns using the GA. Then it generates the desired behavior corresponding to various landmarks. The results of the algorithm is promising and has a high accuracy in classifying the input patterns. The effectiveness of the developed system is demonstrated by simulation and experimental studies.