Drone Visual Positioning Algorithm with Convolutional Neural Network

Since the Global Positioning System (GPS) may not always be precise and available, it is better to adopt the vision-based approach in the positioning of a drone. The geometric features are determined to position the drone and find the flying path. In this study, we designed a shallow network with the concept of data redundancy to develop a visual positioning algorithm by eliminating the disadvantages of using GPS and implementing the embedded system. Finally, an ortho-photomap was used to verify the algorithm, and statistics were collected on critical factors to show the performance of the algorithm. The relationship between the amount of redundant data and the accuracy of our algorithm was used to trade off the computational cost and the system performance.