Graphics Reference
In-Depth Information
The experience was often frustratingly slow transmission of the large image files,
which resulted in a continuously growing backlog of imagery waiting to be transmit-
ted. The ground station computer that processed (or merely displayed) the imagery
was often waiting while the UAV continued to collect more imagery.
9.2 Embedded System Overview
The embedded system consisted of hardware and software components, some COTS
and some custom-built. Only the tight coupling of these components permitted effi-
cient and fault-tolerant operation in face of large amounts of imagery data, inter-
mittent and weak connectivity, and low-latency requirements for getting crucial
information to the operators.
9.2.1 System Architecture
The ground control station (GCS) desktop computer (Windows XP, Intel Core 2
Quad Core, 2.40GHz, 3GB RAM) was connected to the UAV via a Wave Relay
1
mobile ad hoc network, specifically, through several low-gain sector antennae on
the ground (approx. 9dBi, 2.3-2.5GHz, 600mW) and low-gain (approx. 1.8dBi),
omnidirectional antennae on the aircraft. The UAV was a customized Sig Rascal 110
ARF remote-controlled aircraft, shown in Fig.
9.1
, and maintained by the Center for
Autonomous Vehicle Research (CAVR) at the Naval Postgraduate School. A separate
VHF control and navigation link were dedicated to the Piccolo flight control system.
Prior to this configuration, the compute-payload consisted of two PC-104 boards
(ADL MSM800XEV, 500MHz AMD processor, 256 MB memory), where one PC-
104 board was dedicated to flight and navigation tasks, and the other performed all
camera operations [
2
]. Careful requirements analysis suggested that a single-board
solution is more capable yet consumes less battery than two boards (see Sect.
9.5.4
).
Some services had to be ported from Windows to a Linux OS.
Hence, the payload consisted of a powerful PC-104 board (Advanced Digital
Logic ADL945PC-T7400 with Intel Core 2 Duo, 2.16GHz, 4M cache) and a Cannon
G9 PowerShot still image (12MP) camera, mounted on a custom-built 2-DOF gimbal
(two degrees of freedom: pitch and roll). The gimbal pointed the camera straight down
irrespective of aircraft attitude in order to collect nadir imagery.
Payload capabilities had to be carefully balanced: aircraft jitter, gimbal stabil-
ity, and image sensor sensitivity determined exposure time and image resolution.
Higher resolution is advantageous for detection, but also requires more memory and
processing resources, and in turn more battery, more weight, more heat dissipation,
and reduced flight time. The chosen boards were selected for their sweet spot in terms
