Objective
Evaluate and validate the onboard landing pad detection pipeline using the Arducam UC788 camera on Raspberry Pi, measuring detection accuracy and latency to confirm readiness for integration into the autonomous landing workflow.
Background / Context
Reliable landing pad detection is a safety-critical capability for the Sky Warriors drone. Once the Arducam UC788 is confirmed operational (#arducam-setup), this issue covers evaluating detection performance end-to-end — from raw camera frames to confirmed pad localisation — under representative real-world conditions.
Technical Tasks
Define landing pad design (colour, shape, marker — e.g., ArUco, AprilTag, coloured circle)
Set up physical test environment with landing pad at various distances and orientations
Integrate or adapt detection pipeline (OpenCV / custom model) with live Arducam feed
Implement ROS 2 node or standalone script to process frames and publish detections
Run detection at multiple distances (e.g., 0.5 m, 1 m, 2 m, 5 m) and record results
Record detection accuracy (true positive rate, false positive rate) per distance
Measure end-to-end detection latency (camera capture → detection output)
Test under varying lighting conditions (indoor, outdoor, low light)
Test under varying approach angles (0°, 15°, 30° tilt)
Log results in a structured format (CSV or JSON)
Document failure modes and edge cases observed
Acceptance Criteria
Detection accuracy ≥ 90% true positive rate at ≤2 m distance
End-to-end detection latency ≤ 100 ms per frame
Detection pipeline runs stably on Raspberry Pi without memory/CPU overload
Results documented in docs/testing/landing_pad_detection_results.md
Dependencies / Blockers
Depends on: Arducam UC788 & Raspberry Pi Setup (Issue #1)
Landing pad physical marker must be printed/fabricated
Detection algorithm or model must be selected prior to testing
Definition of Done
All test scenarios executed and results logged
Acceptance criteria met and verified with reproducible test runs
Results report committed to repository
Findings summarised in this issue as a comment
Objective
Evaluate and validate the onboard landing pad detection pipeline using the Arducam UC788 camera on Raspberry Pi, measuring detection accuracy and latency to confirm readiness for integration into the autonomous landing workflow.
Background / Context
Reliable landing pad detection is a safety-critical capability for the Sky Warriors drone. Once the Arducam UC788 is confirmed operational (#arducam-setup), this issue covers evaluating detection performance end-to-end — from raw camera frames to confirmed pad localisation — under representative real-world conditions.
Technical Tasks
Define landing pad design (colour, shape, marker — e.g., ArUco, AprilTag, coloured circle)
Set up physical test environment with landing pad at various distances and orientations
Integrate or adapt detection pipeline (OpenCV / custom model) with live Arducam feed
Implement ROS 2 node or standalone script to process frames and publish detections
Run detection at multiple distances (e.g., 0.5 m, 1 m, 2 m, 5 m) and record results
Record detection accuracy (true positive rate, false positive rate) per distance
Measure end-to-end detection latency (camera capture → detection output)
Test under varying lighting conditions (indoor, outdoor, low light)
Test under varying approach angles (0°, 15°, 30° tilt)
Log results in a structured format (CSV or JSON)
Document failure modes and edge cases observed
Acceptance Criteria
Detection accuracy ≥ 90% true positive rate at ≤2 m distance
End-to-end detection latency ≤ 100 ms per frame
Detection pipeline runs stably on Raspberry Pi without memory/CPU overload
Results documented in docs/testing/landing_pad_detection_results.md
Dependencies / Blockers
Depends on: Arducam UC788 & Raspberry Pi Setup (Issue #1)
Landing pad physical marker must be printed/fabricated
Detection algorithm or model must be selected prior to testing
Definition of Done
All test scenarios executed and results logged
Acceptance criteria met and verified with reproducible test runs
Results report committed to repository
Findings summarised in this issue as a comment