An adaptive confidence-driven framework for real-time lidar and visual data fusion in autonomous aerial vehicle landing site assessment

Drones and Aerial Vehicles are now increasingly relying on autonomous landing site assessment to find a safe landing spot. The modern day Technique for finding safe landing site involves use of Both LiDAR and Image data, LiDAR data is used for detecting slope, uneven surfaces, and obstacles like rocks, trees, poles, or buildings based on their 3D structure and image data is used for identifying texture and material understanding. For example - A flat water surface may look safe as per LiDAR data due to its flatness but only through image data we can identify that it is a water surface. Hence it is important to assess safe landing site based on both LiDAR data and image data. The current approach seems to be perfect for static environments but how about in the case of dynamic environments where a safe landing spot identified at a particular time might no longer be safe due to some new obstacle appearance at that spot. This paper addresses this issue by proposing an Adaptive confidence driven algorithm that helps in finding safe landing spots in dynamic or changing environments.