Multirole UAVs Supported Parking Surveillance System

The paper deals with the possible application of Unmanned Aerial Vehicles (UAVs) in recognizing irregular traffic situations. This primarily refers to improperly parked vehicles. Fleets of drones are used, and their roles can be observers or inspectors, where each of the roles has specially defined tasks. A case study was conducted in a real urban environment. Traveling salesman problem and Vehicle Routing Problem algorithms were used to determine drone trajectories. The study assumes difficulties in collecting a sufficient number of images for successful training of Convolutional Neural Network, so a synthesized training set was used. This process assumes the use of object detection algorithms. Accuracy was tested on a small test set of real-life images. It has been shown that by moving some of the real-life images from the test set to the training set, the accuracy can be increased. In addition, the paper describes the architecture of the Smart UAV Monitoring System for Parking Supervision (SmartUAV-PS), and tests its functioning in real conditions.