In this paper, we propose a feature-density based path planning algorithm for a quadcopter to execute a Return to Home (RTH) task in GPS-denied environments. The algorithm enhances the accuracy of VINS(Visual-Inertial Navigation System) by employing feature-rich trajectory when satellite signals are jammed or interrupted. In challenging scenarios like dense forests, urban canyons, or underground areas where visual landmarks may be sparse, our method dynamically adjusts the drone's trajectory to maximize the traversal of visually rich areas. By utilizing DBSCAN (Density-Based Spatial Clustering of Applications with Noise), we identify feature-rich regions with high feature density and abundant visual information, so that include them into the planned route. The outlier removal techniques are used to avoid inefficient detours that do not contribute to visual-based navigation, further optimizing the path. A weighted vector summation strategy is applied to balance between the shortest RTH route and maximizing the use of these high-density feature regions to enhance localization accuracy. Through simulation in environments with varying feature densities, our results demonstrate that the proposed algorithm avoids visually sparse areas while maintaining efficient path planning. The drone's ability to adapt its route to the environment improves overall navigation stability and ensures accurate localization even in feature-sparse conditions.
原文链接
分享
引用
