KAZE features-based SLAM for autonomous UAV navigation in GPS-denied environments

Conventional Unmanned Aerial Vehicles (UAVs) heavily depend on the Global Positioning System (GPS) for performing navigation operations. However, due to the high uncertainty of GPS signals in indoor environments, dense forests, and underground locations, it becomes difficult to navigate further. Moreover, GPS signals can be deliberately disrupted in situations like jamming and spoofing. Several approaches have been proposed such as Landmark, Star Tracker, Simultaneous Localization and Mapping (SLAM), and Inertial Measurement Unit (IMU) based methods to address the challenge of navigating in GPS-denied areas. Among these techniques, this study employs Feature-based SLAM, a subcategory of Visual SLAM. Feature-based SLAM is considered to be a reliable method for autonomous navigation of UAVs due to its invariance to rotation, translation, illumination, and scale. Despite this, lack of feature correspondence, high computation cost, and time complexity can make the task of navigation arduous for real-time performance. This study addresses these challenges by integrating the KAZE feature detector and descriptor algorithm with the SLAM algorithm. The UAV is equipped with a stereovision system for capturing the stereo images. The KAZE algorithm operates in a non-linear scale-space which preserves the boundaries of objects while smoothing noise, providing superior localization accuracy and feature distinctiveness. KAZE features are utilized to simultaneously create the map and localize the UAV in that map while navigating. Thus, the proposed KAZE-SLAM method has the capability to accurately detect highly distinctive and accurate feature correspondence for robust and precise navigation in GPS-denied areas.