T08 / Sketch import
Turn a clear sketch into a checked drawing
- Prepare the source. Photograph a flat sheet in even light. Include the paper corners when perspective correction is needed. Use a PNG, JPEG or WebP under 8 MB and 24 megapixels. Work from a copy and add the source/permission credit.
- Open and orient. Use Open sketch. Rotate the image if needed. North is not inferred from paper orientation.
- Correct perspective if needed. Choose Select four paper corners and click top left, top right, bottom right, bottom left. Alternatively enter those image coordinates. Supply the paper width-to-height ratio, then select Correct perspective. An A4 landscape sheet is approximately 1.414. Do not use this to derive real-world building dimensions.
- Suggest outlines. The threshold separates dark ink from paper. A higher value includes lighter marks. Gap repair expands ink by the selected number of analysis pixels, which may close tiny breaks but also merge nearby lines. Minimum region size excludes small enclosed areas; simplification reduces unnecessary vertices.
- Select and confirm. Inspect each coloured region. Letters, annotation boxes, internal rooms or the paper border may look like a candidate. Tick only the intended building outlines and confirm the review statement. Nothing is imported automatically.
- Correct the drawing. Imported outlines use the existing drawing editor. Select a building and drag a vertex handle, or move the whole object. Rename it, add notes and check against the retained background. Undo reverses the import. Use manual Polygon/Freehand outline if the extraction misses or misrepresents a shape.
- Add survey evidence manually. Position people and cameras, set intended view directions, and use the separate activity view. A sketch arrow is not automatically classified as a bat route or a viewing direction. Confirm north separately.
- Save and inspect. Save the editable JSON and reopen it to check your work. Export SVG/PNG for a report or use Print/PDF. Keep the source sketch and its date/revision. The output is schematic and not to scale.
The tool detects enclosed image regions, not buildings. It does not read handwritten species labels, establish survey coverage, classify behaviour or validate a roost feature.
T09 / Video review
Review a bounded piece of footage
- Keep the original. Open a working MP4 or WebM, no larger than 250 MB. The browser must support its codec. A long source is allowed, but each review window is capped at 60 seconds. Incompatible sources need a compatible working copy; retain original files and metadata.
- Set the window. Enter start and end in seconds from the beginning of the opened source. Apply the window. These are not clock times. Changing a populated window clears its tracks/events after confirmation; save first.
- Name the evidence. Add the project, station/camera and reviewer. Give each target a useful label. Record acoustic/field references in its notes. The source SHA-256 fingerprint is retained in the project so a different source cannot silently replace it.
- Place a manual seed. Pause on a clear frame. Select Mark/correct a point and click the target. Alternatively enter X and Y between 0 and 1, measured from the top-left image corner, and add the point. Use the time slider and ±1/30-second buttons to inspect adjacent times; these are time seeks, not verified native frame numbers.
- Request suggestions. Choose 5, 10 or 15 samples per second and a search radius. The analysis image is at most 640 pixels wide; the radius is measured in that reduced image. Suggest forward track follows a small fixed image pattern from the latest manual seed at or before the current time. All suggested points remain amber and unreviewed.
- Review failures honestly. The run stops when a match is weak or ambiguous. It can still follow a wrong object with a high match score. Use Next unreviewed point, inspect the source, accept/reject or mark a correction. Rejected points remain in the exported record. Start a new manual seed after a genuine occlusion; do not invent a path across it.
- Accept only what you checked. Enter your reviewer name. Accept individual suggestions, or use the checked-all declaration only after inspecting every pending point. A reviewed point is your observation, not independent software validation. The match score is not a probability of a bat.
- Record an event. Pause at the event, choose an outcome and add the feature/evidence reference and note. Distinguish unresolved, non-bat/artefact and reviewed observations. Each event retains source-relative time. No automatic unique-animal count is generated.
- Save the handover. Export the editable JSON, events CSV and track-point CSV. The latter includes rejected and unreviewed points, their origin and review status. Reopen the matching video before restoring a project. Your original video is not embedded in the JSON.
- Export a visual copy. Export an annotated still or a clip. Clip export replays the selected window at normal speed and records the canvas in real time. It is silent and re-encoded, with source times and a review notice. Keep the tab visible; leaving it cancels export. Inspect playback, timing and overlays before use. Use the original for audio and primary evidence. Some WebM players initially show an unknown duration; check playback and the source-time overlay rather than relying on that duration display.
Know where these tools can fail
Sketches
Faint or broken strokes, shadows, curled paper, textured backgrounds and overlapping arrows can prevent extraction. Perspective correction assumes a flat sheet. Gap repair changes boundaries by a few image pixels. Manual correction remains essential.
Video
Small targets, compression, motion blur, foliage, rain, insects, crossing targets, camera movement and changing thermal contrast can lose or confuse the tracker. Sampling skips source frames. A track may stop early or follow the wrong target.
Ecological interpretation
An image trajectory alone does not demonstrate emergence, species, abundance, absence or adequate survey coverage. Check the source, context, field notes, audio and limitations. These tools do not replace competent survey design and review.
Local processing and device limits
No survey content is uploaded or automatically saved. Closing the tab can lose unsaved work. Processing and exported media use your device's memory and CPU. Smaller working clips are easier to review. Mobile supports review and controls; a desktop is preferable for precise tracing and video annotation.
This is a functional pilot, tested with labelled synthetic examples. Performance on ACP's real infrared/thermal footage still needs an expert acceptance trial. Synthetic success is not field validation.
Your review checklist
Sketch acceptance
- Does the imported shape match your source after correction?
- Can you reject text/boxes and recover missed shapes by tracing?
- Can you move corners, undo and reopen the project?
- Do the exported views retain labels, source credit and scale limits?
Video acceptance
- Can you find the original observation at the exported timestamp?
- Are wrong matches, failures and corrections visible?
- Do saved projects reopen only against the matching source?
- Are the still and clip readable, correctly labelled and useful?
Record false positives, missed targets, drift, time spent reviewing and any unusable exports. Use those observations to decide whether an output is suitable for your evidence and reporting needs.
Basis, sources and ownership
BCT's NVA technical-guide overview emphasises equipment limitations, data management and fieldcraft. These tools support a review record; they do not claim BCT approval or implement the full paid technical guide.
- BCT: Bat Emergence Surveys with Night Vision Aids — technical guide overview
- MDN: canvas captureStream and MediaRecorder codec support
- MDN: video-frame callbacks and media timing
- W3C: media capture from DOM elements
Checked 17 November 2026. The sketch method uses thresholding, connected enclosed regions, polygon simplification and a four-point projective transform. The video pilot uses normalised cross-correlation of a user-selected image patch. No trained bat classifier is used.
Original ACP interface, drawing templates and test fixtures carry ACP credit. User footage, sketches and third-party imagery retain their own rights. You may use these tools for your own survey workflow and edit your own records. Retain ACP credit and version information on templates and overlays. This does not grant rights to third-party material or assign your source content to ACP.
Practical questions
Why process in the browser for this first version?
It allows local processing without uploading footage or reserving ACP server processing. A visitor could inspect browser-delivered code. Keeping computation proprietary on the server requires the separate hosting, quota and data-management decision in the saved plan.
What comes after this pilot?
Test with representative ACP footage and sketches, fix the measured failure cases and review the outputs. The later five-minute hosted extension needs cost, retention, queue and access-limit evidence before deployment.