Vision systems / Technical proposal / Nov 2026

Vision to motion.

Camera tracking. Shared position. Controlled launch.

A concept for a GPS-referenced smart camera, an orientable launchpad, and a DIY racing-style drone. Demonstrate the complete observation-to-launch sequence with a cooperative target and non-contact flight.

SMART VISIONGPS + CAMERA POSEHUMAN-SUPERVISED
Interactive field demonstrator
Synthetic scene · Not to scale
Camera and launchpad animation. Use the stage buttons below for an accessible explanation of each phase.
01 / Acquire the cooperative targetC: CAMERA ANCHOR   P: PAD ANCHOR
C & P = independent GNSS references
Scene coordinates are illustrative only
Cooperative targetDemo droneViewing direction / FOVCamera FOV
00.0 / 32 s
Printed snapshot. Open the HTML file to play the camera, pad and drone animations.
Demonstration boundary: the camera follows the target; the pad points into a pre-cleared flight corridor. Takeoff and a separate observation route illustrate the response. Physical interception, target-directed launch, and pursuit guidance are outside this proposal.32-SECOND CONCEPT SEQUENCE

Ready. Acquire phase. Cooperative range reference scenario.

01 / The proposition

Prove perception and coordination first.

Build a controlled research demonstrator that connects visual tracking, a location estimate with explicit uncertainty, and an operator-approved launch sequence. The target is known and cooperative throughout the experiment.

01 — SEE

An eye that follows.

Use a suitable smart camera and vision-processing software to detect the test target and maintain its identity. A pan/tilt mount keeps it in view while recording the camera’s measured orientation.

02 — LOCATE

A position with context.

Combine the camera’s GNSS reference, calibrated viewing direction, and validated range evidence. Report a region of uncertainty and data age; retain bearing-only status when range is unknown.

03 — DEMONSTRATE

A supervised response.

Use the pad’s independent GNSS and measured orientation to register it in the same scene. Demonstrate pad rotation, then a controlled takeoff into a separate observation corridor.

THE KEY FEASIBILITY QUESTION

Can the vision system maintain a trustworthy observation at the intended test distance? Reliable tracking and depth must be measured before any launch sequence is connected to hardware.

02 / System architecture

Four modules. One shared picture.

The camera and pad each need a position reference and a measured orientation. GNSS provides position; camera calibration, mount feedback, and pose sensing provide the viewing and pad directions.

CVISION STATION

Observe

Smart camera, detection model, persistent track ID, pan/tilt feedback, and an external GNSS reference. Confirm the chosen camera’s sensors and host requirements.

OUTPUT / image observations
+ timestamp + measured pose
EOBSERVATION HOST

Interpret

Associate the observations in one coordinate frame. Keep bearing-only and range-supported estimates distinct. Display uncertainty, track continuity, and stale-data warnings.

OUTPUT / observation record
+ quality + data age
PPAD DEMONSTRATOR

Orient

Register the pad independently. Show rotation toward an operator-selected, pre-cleared corridor on a bench mockup, followed by a separately approved takeoff demonstration.

OUTPUT / orientation feedback
+ readiness + operator decision
DDEMO AIRCRAFT

Depart

A DIY racing-style airframe with GNSS and a flight stack suitable for stabilized navigation. Fly a separated test route under pilot supervision and record flight telemetry.

OUTPUT / flight state
+ telemetry + test recording

Shared foundation: time synchronization · calibration records · consistent position and height references · operator supervision · event logs

03 / Technical feasibility

GPS is an anchor. It is not depth.

A target can appear at the same image location at many different distances. Camera GPS plus a pixel observation cannot uniquely determine the target’s 3D position.

What the prototype must establish

Camera position
An external GNSS receiver or a surveyed reference locates the camera. The selected camera should not be assumed to include GNSS.
Viewing direction
Calibrated optics and measured camera pose establish the viewing direction. Single-antenna GPS position alone does not provide camera attitude.
Range evidence
Start with an independently measured, cooperative reference. Evaluate stereo depth only where the chosen camera and scene produce usable measurements.
Common reference
Camera and pad observations must use compatible coordinate, altitude, and timing references. Unsynchronized or stale observations must be visibly flagged.
Airframe suitability
A racing drone’s GPS rescue feature is not proof of full navigation capability. Confirm the flight controller’s supported modes before selecting the demonstrator.
04 / Integration sequence

Integrate in four verified stages.

Progress from recorded observations to a ground mockup, then to supervised flight. Each stage must produce reproducible evidence before the next begins.

Stage 01 / Define

Simulation & scope

Agree the cooperative target, observation conditions, flight boundaries, and reporting format. Review this animated sequence with the test team.

GATE 01Test plan and success criteria agreed.
Stage 02 / Observe

Camera bench study

Capture representative footage. Evaluate target identity, pan/tilt tracking, pose records, and range availability against a known reference.

GATE 02Measured errors and failure cases documented.
Stage 03 / Coordinate

Ground demonstration

Display camera and pad references together. Use a non-flying pad mockup to demonstrate orientation, readiness feedback, and held-launch scenarios.

GATE 03Operator control and hold states demonstrated.
Stage 04 / Validate

Supervised flight

Demonstrate takeoff and a separate observation route with pilot supervision. Compare logs with reference data and produce a feasibility report.

GATE 04Evidence supports a go / revise / stop decision.
05 / Interfaces & validation

Make every observation traceable.

The prototype should preserve the origin, time, and validity of each observation. These records support test review and uncertainty assessment; they are not aircraft navigation commands.

C → E

Camera to observation host

Image timestamp, persistent track identity, image observation, detection score, measured camera pose, and calibration reference.

Verify

Track continuity against labelled footage. Distinguish missed detections, occlusions, and identity switches.

R → E

Reference to observation host

Camera position reference, range provenance, reference timestamp, and an explicit indication of unavailable or invalid measurements.

Verify

Compare reported observations with an independent cooperative reference. Report range availability and uncertainty separately.

P → UI

Pad to operator display

Pad position reference, measured orientation, parked / moving / aligned status, and the operator-selected corridor identifier.

Verify

Show requested and measured orientation as separate states. A visual alignment animation is not hardware position feedback.

D → LOG

Aircraft to test record

Pilot approval event, flight state, telemetry timestamp, and the recorded takeoff and separate observation-route sequence.

Verify

Review the log and video together. Confirm the hold scenarios prevent departure and the flight remains in its cleared corridor.

Observation states

Observed. Qualified. Stale.

Distinguish image detection, bearing-only tracking, a range-supported estimate, and an invalid or stale record. Never convert detection confidence into a claim of position accuracy.

Bench acceptance

Test the missing-data cases.

Exercise absent range, track loss, delayed pose, stale timestamps, and missing operator approval before flight. Agree measurable acceptance criteria and test conditions in advance.

06 / Technical deliverables

Validate the complete
observation chain.

Record the evidence needed to assess perception and launch coordination. Set pass/fail criteria before testing, then assess track continuity, range availability, error against the cooperative reference, observation age, and correct hold-state behaviour.

Camera, processing software, and interface selection will be confirmed against prototype requirements.

  • Annotated architecture and reference-frame assumptions
  • Camera footage, track logs, and calibration records
  • Range and position uncertainty assessment
  • Pad motion and held-launch demonstration
  • Supervised takeoff and separate-route recording
  • Feasibility report with go / revise / stop recommendation