부산대학교 도서관

Global Positioning System (GPS) is becoming necessary in today's life. However, if GPS signals are interfered, blocked or missing, the situations will result in significant impacts in navigation causing unforeseen consequences. This paper proposes an image-based GPS-less positioning system which is realized on the basis of the SIFT algorithm, FLANN matcher and YOLO neural network model. The system compares the scene features with feature images from the pre-established feature image database and obtains the corresponding GPS coordinates from the image with the highest matching degree. It fuses GPS coordinates from image recognition and inertial measurement unit (IMU) data through Kalman filter. Preliminarily experimental results are presented.