Alternative navigation method for unmanned aerial vehicles in global positioning system denied environments

Navigation system for both unmanned and manned aerial vehicles are affected negatively in Global Positioning System (GPS) denied environment. Although pilot can take control on navigation in GPS-denied environment for manned aerial vehicles, for unmanned aerial vehicles (UAV), GPS-denied environment can cause catastrohic solutions. Because of this reason, it is one of the most common problems for aircrafts to continue navigation in GPS denied environment. In the case of GPS loss, even if the aircraft can continue its navigation for a while with the inertial navigation system (INS), the error due to the drift increases and the aircarft is moving away from the route it should follow. For this reason, a system that does not feed from GPS information has been developed and it enables the UAV to know the location of the aircraft with a payload camera placed on the UAV. There are two types of image data, one of which is from the payload camera placed on the aircraft, and the other is satellite image set which has GPS information for each satellite image. The position of the UAV is determined by using the feature extraction neural network model and normalized variant mutual information within two image sets. Images from the satellite image set are labelled with GPS information and given to the UAV. After these two sets of images are passed through the different feature extraction neural network models and classification and feature extraction is done for the image datasets, these features are passed through two more designed neural network. After passing on the first neural network, it is determined whether the satellite images are in the images coming from the UAV or not, and if there are satellite images in the image coming from the UAV, the second neural network is passed through and the UAV location is matched over the satellite images. Along with the results from the second neural network, normalized variant mutual information is used within the two image sets. In order to find the region where the match is maximum, maximum of normalized variant mutual information between these two images is calculated by giving more weight of the results from second network. In case of GPS loss, the developed algorithm starts to work and satellite images which have GPS information are searched in UAV camera images. If the UAV is in that area, GPS information from satellite image is given to the UAV. To test our algorithm, we find two image dataset, where one consists of images taken from UAV and the other one consists of the satellite images, and labelled these images. We conduct various experiments under different scenarios such as different feature extractor networks, stitched/unstitched input images on the UAV side. We observe that the performance after the mutual information module is higher compared to the performance of the neural network in terms of both mean squared error and intersection of union.