03 · SCHOLARLY
304 FILES·LAST 5D AGO
← Scholarly / Papers
SCHOLARLY / PAPER

A Hardware and Software Platform for Aerial Object Localization

Matthew Szenher · Alex Delacroix · Eric Keto · Sarah Little · Mitch Randall · Wesley Andres Watters · Eric Masson · Richard Cloete

Journal of Astronomical Instrumentation · 2023

The Galileo Project's first observatory node at Harvard deploys a weatherized nine-camera array spanning LWIR, NIR, and visible wavelengths, with a novel ADS-B-based extrinsic calibration algorithm and stereo triangulation software capable of reconstructing full 3D kinematics of aerial objects.

SharePostReddit

Brief

Szenher et al. (2023) document the hardware and software stack for the Galileo Project's initial UAP observatory at the Center for Astrophysics, Cambridge. The primary sensor package, internally named 'the Dalek', is a fiberglass dome housing seven radially arranged FLIR Boson 640 LWIR cameras (50° FOV each, 360° azimuthal coverage, nominally tilted 30° above the horizon) plus zenith IR and NIR cameras; a separate Alcor OMEA 9C provides a 180°×180° visible all-sky view at 9575×6380 resolution. A novel extrinsic calibration routine exploits FAA-mandated ADS-B transponder data from overhead aircraft to align camera orientation without manual ground targets. Preliminary thermal detection range tests place a 60 m airliner at a maximum of 35 km and a ~10 m twin-engine propeller plane at 7 km; triangulation error scales with the square of object distance and inversely with stereo baseline length.

Metadata

Category
Hub & Overview
Venue
Journal of Astronomical Instrumentation
Type
Peer-reviewed
Year
2023
Authors
Matthew Szenher, Alex Delacroix, Eric Keto, Sarah Little, Mitch Randall, Wesley Andres Watters, Eric Masson, Richard Cloete
Access
Open access
Length
1.7 M
Programs
Galileo Project
Instruments
FLIR Boson 640 LWIR camera, ZWO ASI462MC CMOS NIR camera, Alcor OMEA 9C visible all-sky camera, NPACKMAN optical instrument, Nvidia Jetson edge computer
Data sources
ADS-B transponder data, GPS/WGS 84
Tags
UAP-physics, instrumentation, aerial-object-tracking, 3D-localization, LWIR-imaging, extrinsic-calibration

Key points

  • Seven FLIR Boson 640 LWIR cameras (7.5–13.5 μm, 640×512 px, 60 Hz) are arranged radially at 30° elevation to provide uninterrupted 360° sky coverage; an eighth Boson with 95° FOV covers the zenith.p.5
  • Preliminary detection-range tests: a 60 m airliner is easily detectable at 20 km and barely detectable at 35 km; a 20 m business jet barely detectable at 23 km; a ~10 m twin-engine propeller plane barely detectable at 7 km, all with the Boson 640.p.5
  • Triangulation distance error scales as Z² and is inversely proportional to the stereo baseline b; accuracy collapses when the two camera look-at angles are nearly equal (parallel geometry).p.4
  • Extrinsic calibration uses a new ADS-B-based algorithm: the camera images overhead ADS-B-equipped aircraft, and the known GPS positions of those aircraft are matched to their image centroids to recover the camera's rotation matrix without any manual target placement.p.8
  • Object detection runs on YOLO per camera on an Nvidia Jetson edge computer; DeepSORT links detections across frames to form tracks; centroid detections are transmitted over 0MQ to a central Thinkmate processor for 3D localization.p.3
  • The Dalek dome is structurally rated to survive 200 mph winds, above Category 5 hurricane threshold (157 mph), with FEA showing less than 1 mm radial deformation at that load.p.6
  • The Alcor OMEA 9C all-sky camera covers 180°×180° at 350–750 nm, with exposure times from 32 μs to 1 hour and a resolution of 9575×6380 pixels; a separate NPACKMAN instrument adds a 170°×170° optical camera for space-weather monitoring.p.6
  • Timing synchronization between cameras is treated as a non-trivial engineering problem: unsynchronized frames introduce localization error proportional to v·Δt/b, where v is object velocity and b is baseline; the paper addresses this in Section 4.4 (omitted from provided excerpt).p.4

Verbatim

  • Also, infrared data can be used to detect and characterize exhaust heat (or the lack thereof) from tracked objects.
    p.2
  • Thus, we see that the measurement error in the object distance, Z , due to angle uncertainty, depends on the square of the distance, Z 2 , and is inversely proportional to the baseline distance, b .
    p.4

Most interesting

  • The camera array is nicknamed 'the Dalek', an apparent reference to the Doctor Who antagonist, and is a one-of-a-kind purpose-built hemispherical instrument.
  • If a chessboard intrinsic-calibration target cannot be placed close enough to a camera in the field, the protocol calls for attaching the chessboard to a drone and flying it in front of the camera.
  • Long-wave IR cameras are favored partly because detection depends on object surface emission rather than reflected sunlight, making them effective at night and independent of solar illumination geometry.
  • The dome enclosure carries eight deployable sunshades on Arduino-controlled servos specifically to prevent direct sunlight from damaging the IR sensors.
  • Two 72-cubic-inch desiccant containers are installed inside the Dalek dome; the desiccant must be periodically baked in an oven to restore its moisture-absorbing capacity.
  • The edge computer enclosure is rated IP65 and shielded against both electromagnetic interference (EMI) and radio frequency interference (RFI), an unusual combination for a UAP-monitoring instrument designed to operate near potential RF sources.

Cross-references