This thesis proposes an attentive landmark recognition scheme for intelligent autonomous navigation systems. The scheme consists of three modules : visual search, recognition, and attention modules.
In the visual search module, the candidate regions which might contain landmarks are determined in an input image, we use an adaptive binarization technique and a bounded visual search method.
From the candidate regions, the feature sets represented by the fuzzy linguistic variables are extracted in the recognition module. And recognition process is performed by a neural network with these features.
For the more efficient visual search, the attention module predicts the position and size of the landmark which was found in previous scene. The prediction is made by a modified neural network equivalent to the Greville's algorithm-a recursive parameter estimation algorithm.
Finally, we evaluate the proposed techniques by experiments.
