Modular interpretation of low altitude aerial images of non-urban environment

In this paper we present a modular algorithm for interpretation of low altitude aerial images of non-urban environment. Non-urban land-covers, e.g., rivers, grass, unlike urban land-covers, have naturally unstructured boundaries and are usually containing diverse combination of colour and texture. The proposed method consists of a coarse and computationally efficient module, and a fine interpretation module. The coarse module is able to produce approximate estimations of land-covers using a single colour-base feature and contextual information. In cases when the coarse module fails, the fine module is able to accurately classify the desired land-cover. The fine module uses a combination of boundary, colour, texture and context features for accurate interpretation of the land-covers. The modular method inherits the high accuracy from the fine module and low computational expense from the coarse interpretation module. Experimental results show that the proposed algorithm can detect the target land-covers in low altitude aerial images of non-urban environment with acceptable accuracy and low computational requirements.