The Construction of Environment Models with the Aerial Images Captured by Low Altitude

Abstract

There are many difficult technical issues are remained in the automatic low altitude flying task of an UAV (Unmanned Aerial Vehicle), such as environment modeling and obstacle avoidance. Solving such issues will enable the further application of low altitude UAV in the fields of emergency response in urban areas, access to disaster, etc. This paper proposed a method for constructing the urban environment models from the aerial images captured by the UAV. Although there are many kinds of objects in the urban environment, we focus on setting up a description model for the classic cubic buildings which appear most frequently in the city scene. In this method, such cubic urban buildings are described by an expansion of the well-known HOG (Histogram of Oriented Gradient) descriptor which can represent the object appearance with the statistical histogram. Based on the truth that the urban buildings usually contain massive well-organized parallel gradients that are orthogonal to each other, the environment model for describing urban buildings could be expressed by its orthogonal property in HOG, where the interval between the first and second histogram peaks obtained by the Mean Shift Algorithm should be near to 90 degrees. An object with such orthogonal property in HOG feature will considered as a classic urban building, otherwise penalized as the background component. Through extensive experiments, the efficiency and effectiveness of proposed algorithm have been confirmed.