Optical Flight Evidence Scanner · Method 2.1 · 14 October 2026
A trustless smartphone observation and optical trajectory method for mosquito mapping
Turning an ordinary phone camera into a reproducible field scanner and an AQI-like evidence network for mosquito presence
Abstract
The Optical Flight Evidence Scanner converts a participant's mosquito count and a three-second smartphone video into one geotagged, independently checkable field record. The participant states how many mosquitoes are visible in the framed area; the source clip, SHA-256 digest, timestamp and GPS fix preserve the evidence for that statement. In parallel, every encoded frame is stabilized and compared with a temporal background. Adaptive residual analysis proposes small moving targets, an analysis agent checks time-labelled source crops and supported fragments are consolidated to prevent duplicate inflation. Each point can then be enriched with breeding-habitat and building-adjusted exposure indicators for an AQI-like mosquito presence map.
1 · Key terms
A scanner result is built from evidence, not a single bright pixel.
| Term | Meaning |
|---|---|
| Observer-stated count | The number of mosquitoes the participant reports seeing in the framed area at capture. It is stored as the field observation and is not silently replaced by the optical result. |
| Proof package | The stated count linked to the complete three-second clip, SHA-256 file identity, capture timestamp, GPS point and analysis version. |
| Candidate peak | A localized change retained by the adaptive motion stage for possible tracking. |
| Tracklet | A time-ordered motion hypothesis across source frames. Tracklets may overlap and are not counts. |
| Supported trajectory | A tracklet with coherent motion and agent-supported observations in the original video frames. |
| Trajectory group | Supported fragments consolidated into one observed passage to prevent duplicate counting. |
| Maximum observed together | The largest number of supported groups overlapping in time inside the recorded view. |
| Visual identity verdict | A separate assessment of whether visible morphology is mosquito-like; it does not alter the retained path evidence. |
2 · Prior-art alignment
Commodity capture, trajectory science and automated image evidence in one field workflow.
| Prior work | Established capability | Method contribution |
|---|---|---|
| Phone acoustic surveillance [2] | Commodity phones can contribute to mosquito surveillance, although useful acoustic range is constrained. | The scanner uses a participant-directed camera view and does not depend on wingbeat audio. |
| Citizen-science imaging [3] | Phone images can support automated mosquito identification when visual detail is available. | Short video adds movement, time and path evidence before visual classification. |
| Flight reconstruction [4] | Mosquito behaviour can be studied from reconstructed trajectories. | The workflow brings image-plane trajectory evidence to an ordinary browser phone without specialist multi-camera hardware. |
| Autonomous sentinels [5] | Optical triggering and automated analysis can reduce manual surveillance effort. | A geotagged, tamper-evident web workflow makes the observation portable across participating phones. |
3 · Capture and computation
Point, state, record and preserve the complete evidence path.
- Point.The participant directs the rear camera toward visible mosquito flight activity. The browser selects the best camera mode the device can provide.
- State.The participant enters the number of mosquitoes visible in the framed area. This direct observation becomes the point-level field count.
- Geotag.Location and accuracy are captured with the observation, providing a point record for GIS use.
- Record and bind.A three-second clip is recorded and linked to the stated count. The complete clip is uploaded with a SHA-256 content hash so its identity can be checked throughout analysis.
- Stabilize.All encoded frames and real timestamps are decoded. ORB/RANSAC alignment estimates camera movement and preserves mapping back to original-camera coordinates.
- Scan.A temporal background, local-noise estimate and texture mask expose compact airborne change while suppressing global illumination and static structure.
- Track.Gap-tolerant competing hypotheses connect candidate peaks through time. Alternatives remain separate until the evidence supports consolidation.
- Review.The analysis agent receives original-camera crops alongside the enhanced motion scan and identifies which exact frame observations support a coherent airborne trajectory.
- Consolidate.Overlapping or adjacent fragments of the same passage are merged. The final record reports supported trajectory groups and the maximum number observed together.
- Map.The observed count and video proof remain distinct from the automated trajectory evidence, then join breeding-habitat and building-adjusted human-exposure indicators at the capture location.
4 · Evidence rule
The enhanced scan proposes; original video frames verify.
Stabilized change
Compact change survives camera-motion and broad-lighting suppression.
Time-consistent path
Locations form a plausible airborne path across real frame timestamps.
Visual support
The agent confirms the proposed observations in their original-source crops.
support = S ∧ T ∧ VA bright residual by itself is insufficient. The retained result links each supported path point to its source frame and timestamp.
The participant's stated count and automated trajectory result are recorded side by side. Keeping them separate prevents circular confirmation: the count documents the in-situ observation, while the optical scan supplies an independent, replayable evidence layer.
5 · Field trajectory frames
From candidate motion to filtered trajectory evidence.
These field examples show how the algorithm handles artifacts and duplicate paths in ordinary phone video. The coloured overlays preserve the candidates before final evidence filtering: a line represents a proposed tracklet, not a mosquito count. The timestamped sequences show the original-camera crops and enhanced scanner responses used to evaluate those proposals.
Suppress background change
Camera alignment, illumination correction and local-noise estimates suppress broad image changes and persistent scene texture.
Check the original frames
Time-consistent paths are checked by the analysis agent against original crops. Artifact and indeterminate verdicts are excluded.
Consolidate fragments
The finalizer joins supported fragments with shared observations or compatible timing and positions into trajectory groups.
14 candidate tracklets → 9 artifacts and 1 indeterminate excluded → 4 supported fragments → 2 trajectory groups. The final groups were observed simultaneously in the early frames.
Before final filtering · 14 candidate tracklets
6 candidate tracklets → 2 indeterminate excluded → 4 supported fragments → 1 trajectory group. Shared observations connect the retained fragments, so the finalizer consolidates them into one group and flags association uncertainty.
Before final filtering · 6 candidate tracklets
How to read these panels: the yellow circle locates the candidate. The upper tile shows its original appearance and the lower tile shows enhanced image change. Review retains only supported frame observations: for A1, frames 1–3 are retained and frames 4–5 are excluded. These examples demonstrate motion filtering and duplicate consolidation; their visual identity verdict remains unresolved.
6 · Duplicate-resistant counting
Count observed passages once, even when the detector proposes several fragments.
Short, overlapping tracklets deliberately give the system several ways to recover faint movement. The finalizer then joins temporally compatible fragments that describe the same path. Concurrent, spatially distinct paths remain separate. The operational optical unit is therefore a supported trajectory group: one supported passage through the camera view.
The participant-stated mosquito count remains the field measurement. Trajectory grouping strengthens that measurement by showing what moved, when it moved and which proposed fragments were consolidated instead of adding duplicate evidence.
7 · Trust and provenance
Trustless evidence means the record can be checked without relying on one assertion.
The observer-stated count is bound to the uploaded clip, and the clip, extraction configuration and agent request each receive a SHA-256 digest. Exact timestamps, source coordinates, GPS accuracy, quality flags, model identity and software versions accompany the result. A failed or incomplete analysis cannot silently become a successful zero. Re-analysis creates a new versioned record instead of rewriting the evidence. This produces a transparent chain from what the participant reported, to what the camera recorded, to what the algorithm retained, and finally to where the observation appears on the map.
8 · Field replay evidence
Frame-level replay exercises faint motion, simultaneous paths and duplicate consolidation.
Three field clips were replayed through the complete version-2 workflow. The automated output contained zero, two and one supported trajectory groups. Against five independently annotated trajectory passages, the updated extractor recovered three, compared with one under the previous extractor. The replay demonstrated reproducible end-to-end operation across faint motion, simultaneous paths and fragment consolidation, with every retained result linked back to its exact source-frame evidence.
Multi-site deployment will combine observer-stated counts, replayable video evidence, trajectory agreement and device-stratified quality with habitat and built-environment context. The resulting point network is designed to support an AQI-like mosquito presence view for community surveillance, hotspot prioritisation and targeted field action.
References
Scientific foundations
- 1
McMeniman CJ, Corfas RA, Matthews BJ, Ritchie SA, Vosshall LB. Multimodal integration of carbon dioxide and other sensory cues drives mosquito attraction to humans. Cell. 2014;156:1060–1071. DOI / source ↗
- 2
Mukundarajan H, Hol FJH, Castillo EA, Newby C, Prakash M. Using mobile phones as acoustic sensors for high-throughput mosquito surveillance. eLife. 2017;6:e27854. DOI / source ↗
- 3
Pataki BA, Garriga J, Eritja R, Palmer JRB, Bartumeus F, Csabai I. Deep learning identification for citizen science surveillance of tiger mosquitoes. Scientific Reports. 2021;11:4718. DOI / source ↗
- 4
Butail S, Manoukis NC, Diallo M, Ribeiro JMC, Lehmann T, Paley DA. Reconstructing the flight kinematics of swarming and mating in wild mosquitoes. Journal of The Royal Society Interface. 2012;9:2624–2638. DOI / source ↗
- 5
Lin N, Qian Y, Wei L, et al. An autonomous intelligent mosquito sentinel for field-deployed surveillance. Communications Engineering. 2026. DOI / source ↗