Emergency Landing Field Identification (ELFI)
ELA requires geo-data of at least one or better several runways which are suitable for the considered aircraft. For these runways ELA determines whether and how these are reachable from the emergency position by gliding. A runway should be preferably aligned into the wind. Additionally, it has be long and width enough in the direction of landing to ensure a safe landing with the damaged aircraft. Further requirements concern the slope in longitudinal and transversal direction. Depending on the aircraft type, various constrains regarding to the maximal slope in both directions exist. Generally, runways of published airfields fulfill the mentioned criteria for a variety of different aircraft types like sport aircrafts, airliners, or helicopters. Unfortunately, because of low residual altitude it is sometimes impossible to reach a designated airfield. In this case, an emergency landing field has to be found within reach.
Certainly, an emergency landing field is only the second choice because the surface is commonly not paved and the aircraft may be damaged during the landing. Therefore, the objective is to find an emergency landing field where the probability for a crash landing is as small as possible.
With the ELFI approach, emergency landing fields with slope and dimension requirements specific for an aircraft type can be determined based on LiDAR elevation data. The calculation uses high-resolution elevation data in the range of a few decimeters, which is also available for a meshed grid with edge length under one meter. Thereby, e. g. data in the order of about 70 GByte arises for the city Hagen. Based on this high resolution a huge amount of data must be processed even for small areas to calculate emergency landing fields. To speedup the computations, computer systems with parallel operating multicore processors are applied. This requires a spatial distribution of the elevation data in tiles which are processed concurrently by multiple processor cores and their partial results are merged later. The parallel programming is based on POSIX threads and the data can be processed on an up to date multicore computer in several hours. Thereby, the elevation data is sampled for a limited number of landing directions (e. g. in 22.5 degree steps) and respectively the above mentioned parameters regarding to the size and slope are evaluated. In this way, two runways are found for each landing direction and the edge points are stored in a geo-database.
The identification of emergency landing fields based on LiDAR elevation data is sometimes insufficient. For example, water surfaces are detected as perfect emergency landing fields, although they hide a much higher risk than solid surfaces. Narrow objects like high-voltage power lines or feeders cannot be detected solely with the elevation data. The located emergency landing fields must be additionally processed with corresponding satellite imagery. Thus, besides narrow obstacles also texture of the surface can be analyzed. In an extended ELFI approach satellite imagery was investigated by various standard techniques for image segmentation (not detectable in the elevation data) for obstacles. Currently, we work on the implementation of a semantic classification regarding to the landability by means of specialized artificial neural networks. Thereby, so-called convolutional neural networks (CNNs) are used. This kind of artificial neural networks has proven its worth particularly in the case of visual pattern recognition in the region of autonomous car. Hence, it is expected that the usage of CNNs facilitates a fast and automated pixel classification regarding to the landability of the emergency landing fields which are identified based on elevation data. In contrast to previous CNNs, the artificial neural networks applied in this project should be topological optimized and therefore, better adjusted to the
specific recognition task.
