Inclined Image Recognition for Aerial Mapping by Unmanned Aerial Vehicles

In general, aerial mapping is an image registration problem, i.e., the problem of transforming different sets of images into one coordinate system. Aerial mapping is one of the important capability of an unmanned aerial vehicle (UAV). Here, the images processed by the registration system is strongly influenced by the quality of the image captured by the UAV. To select the image that will be processed efficiently is not easy considering the ground truth in the mapping process is not given before the UAV flies and captures the image. On the other hand, generally, UAV will fly and take the image in sequence regardless of the quality. These will result in several issues, such as: 1) the quality of mapping results becomes bad, and 2) the computational cost of registration process becomes high. To tackle such issues, therefore, we need a recognition system that is able to recognize images that should be excluded from the registration process. In this paper, we define such image as an 'inclined image,' i.e., images captured by UAV not perpendicular with the ground. Although we can calculate the inclination angle using a gyroscope attached to the UAV, our interest here is to recognize the images without the use of such sensor like human do. To realize that, we utilize a deep learning method to build an inclined image recognition system. We tested our proposed system with images captured by UAV. The results showed that the proposed system yielded accuracy rate of 86.4%.