Visual-based Landing Guidance System of UAV with Deep Learning Technique for Environments of Visual-detection Impairment

Most vision-based landing algorithms cannot be applied in a severe vision-detection environment. However, the application of a deep learning technique to vision-based landing algorithm can solve the problem of a severe vision-detection environment, especially in vision-impaired environments. Based on this fact, a novel-landing concept with deep learning technique is proposed in this study. Three main techniques applied for guided landing are 1) deep learning for accurate landing mark detection; 2) location-memorized system for coping temporal failure of the landing mark detection; 3) Unmanned Aerial Vehicle control algorithm for vibration minimization in vision sensor. The proposed system successfully and accurately guided the multicopter onto the landing area without failure in vision-impaired environments. The results show that the proposed landing algorithm can overcome environmental restrictions in operating multicopters.