Drowsy Guard presentation slides¶

A six-slide deck drawn from the research proposal and from what the prototype has actually measured so far. It is deliberately short: concept, flow, equipment, current results, conclusion - one idea per slide, each carried by a picture rather than by bullets.

Slide deck¶

16:9 widescreen, PowerPoint format, with speaker notes on every slide.

Download the deck

Also here: a deck that runs in the browser¶

A second deck, written independently, covers the same ground as eight slides in a single self-contained HTML file - no PowerPoint needed. Arrow keys advance it, Ctrl+P prints it as landscape A4, and it falls back to a drawn build diagram until a real bench photo is dropped in.

Open the browser deck · presenting, exporting and editing it

The two are alternatives rather than halves of one thing - they were built in parallel and overlap almost completely. Pick whichever suits the room, and retire the other rather than maintaining both.

What is on each slide¶

#	Slide	The one point it makes
1	Title	What the device is, who built it, and that it runs with no cloud.
2	Concept	Drowsiness is a duration, not a frame - four signs, each with its threshold.
3	Flow	See → Find → Measure → Confirm → Warn, all five steps on the device.
4	Equipment	Five off-the-shelf parts, about US$ 27 in total.
5	Where it is now	The measured numbers from the first hardware run, and the gap that remains.
6	Conclusion	The effort now goes into training the lightweight on-device model.

Slides¶

Title slide: Drowsy Guard, a low-cost IoT system for driver drowsiness detection — 1 - Title

Concept slide: a drowsy driver with the eye, mouth and head cues marked, beside their duration thresholds — 2 - Concept: eye closure ≥ 1.0 s, long blink > 0.4 s, yawn ≥ 1.2 s, head nod 0.3-1.5 s

Flow slide: five steps from camera capture to a spoken warning — 3 - Flow: See, Find, Measure, Confirm, Warn

Equipment slide: the five parts with an illustration and price for each — 4 - Equipment, priced from Appendix A of the proposal

Results slide: measured frame rates and inference times from the first hardware run — 5 - Where it is now, from the 23 August 2026 hardware run

Conclusion slide: keep the model light, train it in visible light, judge it per driver — 6 - Conclusion and the targets that decide success

Slide 5 has an empty screenshot panel

The dashed panel on the results slide is a deliberate slot for a screenshot of the live preview at 192.168.4.1 - face box, landmarks, risk bar and event log. Capture one from a running board and drop it in before presenting. The numbers beside it are already real and need no change.

Rebuilding the deck¶

The deck is generated, not hand-edited, so a change to a threshold or a measured number is a one-line edit and a rebuild:

cd scripts/presentation
npm install          # pptxgenjs + sharp, first time only
npm run build        # writes docs/assets/documents/DrowsyGuard-Presentation.pptx

gen_assets.js draws every illustration - the component art and cue icons as SVG, the two photographs as crops of docs/assets/images/drowsy-guard-cover.png - into scripts/presentation/build/. build_deck.js then lays out the six slides. Neither the intermediate art nor node_modules/ is tracked; the built .pptx is.

Editing the file directly in PowerPoint works too, but the next rebuild overwrites it - put lasting changes in build_deck.js.