Place Recognition Using a Siamese Neural Network in the Problem of Simultaneous Localization and Mapping with Sparse Sensing

The use of nano-drones to build a map is an effective method of obtaining a plan of hard-to-reach places. The paper presents algorithms for constructing a map using a graph of factors and correcting errors by recognizing a place using a Siamese convolutional neural network. The experimental results obtained with the use of a nano-drone in the autonomous flight mode are presented. The constructed map of the area was evaluated for consistency by a numerical method.