Multi-sensor obstacle detection and tracking for aircraft ground operations

Abstract

Aviation accident reports demonstrate that accidents and incidents during aircraft ground operations have remained unresolved. The majority of these accidents arise from pilot error. Air Traffic Control limit this problem to a certain extent by providing sequencing to ground traffic. However, a proper solution for obstacle avoidance is still inexistent. This thesis addresses the problem of obstacle avoidance by proposing an obstacle detection and tracking technology that can be used to determine the distance of separation between aircraft and obstacles. Unlike previous work, the proposed solution fuses data acquired from two colour cameras and a LIDAR sensor. Image data acquired from the two cameras is fused using stereo vision techniques. These techniques compare pixels between the left and right images to recover depth information. This information is then used to map each image pixel to its corresponding 3D world coordinate. Another set of 3D points is acquired from the LIDAR sensor. The two sets of 3D points, referred to as point clouds, are processed and analysed to detect obstacles. Detected obstacles are passed on to a tracking algorithm that consists of a Particle Filter and an Occupancy Grid. The Particle Filter tracks the positions of detected obstacles whilst the Occupancy Grid tracks their shapes. The tracked information may then be used to determine the distance of separation between the aircraft and obstacles. The proposed technology was evaluated in different scenarios through a series of experiments. The first batch of experiments was carried out in a synthetic environment. Meanwhile, the second batch of experiments was carried out in a real environment. The accuracy and performance of the proposed sensor fusion algorithm were identified. The results show that it successfully detects obstacles and that it manages to improve confidence in the area they occupy.