Visually Assisted Landing System

A CRITICAL NEED

Commercial airliners have been landing automatically for decades, often in little or no visibility.  This ability, however, requires significant equipment both on-board the aircraft, and on the airport.  This has a cost implication for both the airport and the aircraft operator, but the aircraft operator also has the added burden of weight on-board the airplane.  In the world of Unmanned Aircraft, payload (weight of equipment carried) is at a premium, and as a result, most UAVs do not have a similar autoland feature built in.  Some systems use automatic landing at designated facilities, but that again requires the presence of ground equipment.

UAVs require the ability to land autonomously in the event of lost communications, emergencies, or deliberate diversions from planned landing sites.  An automatic landing system can also be used to assist the human pilot during normal, human-flown operations.  Recent significant losses of unmanned systems have shown that there must be a backup plan in the event of failure.  In fact redundancy should carry deeper into the system design with multiple systems integrated to take care of critical flight phases and ensure mission completion and public safety.

To address this need, 2d3 has developed the first phase of a Visually Assisted Automatic Landing System, or VALS.

VALS capitalizes on 25 years of computer vision experience present in the 2d3 team to use an on-board camera to help guide the aircraft to a safe landing, visually.  VALS is the first step toward a completely automatic, visually based navigation system for Unmanned Aerial Systems, allowing a UAV to fly autonomously in the same way a human pilot navigates his airplane - by looking out the window.

THE PROCESS

Automatic Terrain Feature Detection (ATFD)

2d3’s ATFD algorithms initially detect hundreds of points in the image, what we call features.  VALS performs this step automatically, and in realtime with minimal latency, at video frame rates - usually 30 times per second.  This is one of the factors that makes EO camera optics such a robust sensor for determining aircraft position and orientation with reference to ground, it has a high sampling rate.

Feature Tracking and Correlation (FTAC)

The 2d3 FTAC engine now tracks those features across multiple frames to build up a motion model of the scene.  By analyzing the imagery looking for parallax changes, VALS can determine the camera’s motion from the relative motion of the tracked features in the scene.

Calculate Geometry

Using revolutionary computer vision technology, Structure from Motion (SFM), VALS now generates an accurate representation of the 3D geometry of the terrain observed by the onboard camera(s).  We have illustrated this in the image to the left by placing two computer generated arrows at the corners of the runway.  With a prior knowledge of the runway’s dimensions which are taken from the published pilots data rather than any special survey, VALS determines the aircraft’s height above terrain, its distance from the edge lines, and it orientation in the world. 

THE EQUIPMENT

In order to preserve valuable payload space and weight, 2d3 is designing VALS to operate in a very small, very lightweight component that will fit inside all but the smallest, hand launched aircraft.

VALS will use imagery and aircraft state data as inputs, and provide position and orientation data as outputs in a serial data stream for connection to the existing autopilot system.

• Press