Offline Banknote Reader for Blind and Low-Vision Users
A final year project that builds an offline Android app for blind and low-vision people to identify banknotes, count a handful of cash and check change, fully on the device. It targets currencies mainstream accessibility apps do not support, works with TalkBack, and is tested with real visually impaired users.
Intermediate
A small team, or one strong student willing to learn something new.2 semesters · 2 to 3 students
7 to ship
3 optional extrasSuggested stack
What you should ship
- Collected and labelled image set of every current note of one chosen currency, front and back, in varied light, folds and backgrounds
- Small on-device classifier converted to TensorFlow Lite, with a "not a banknote" class to avoid false readings
- Camera screen that speaks the denomination and gives a distinct vibration pattern per note, with no buttons needed
- Counting mode that keeps a running total as notes are shown one by one, and a change check mode against an expected amount
- Full TalkBack support, large controls and high-contrast mode for low-vision users
- Accuracy report per denomination including worn, folded and partially covered notes
- User test with at least 5 blind or low-vision participants, arranged through a local organisation
If you have time left
- Support for a second currency for travellers, switchable by voice
- Detection of old or withdrawn note designs with a spoken warning
- Coin recognition for the most common coins
The problem blind people have with cash
Cash is still used every day in much of the world, and a blind person has to trust whoever hands them change. Some currencies have tactile marks, but these wear off quickly and differ only slightly between notes. Low-vision users struggle with similar colours. Asking a shopkeeper or stranger "which note is this?" many times a day costs independence and invites cheating.
Phones can read notes, but the well-known apps support a short list of currencies and often need a data connection for their broader features.
What already exists and the gap
- Microsoft Seeing AI: a multi-purpose recognition app with currency mode for a set of major currencies, including Indian rupees.
- Google Lookout: a free Android accessibility app whose currency mode supports US dollars, euros and Indian rupees.
- Envision AI: text and scene recognition, subscription based.
- Single-currency student and research apps: several exist on app stores and GitHub, but few are tested with blind users or built for TalkBack.
The gap is a small, offline, well-tested app for a currency the big apps skip (the Pakistani rupee is one example; many African and Asian currencies are others), designed with and tested by the people who will use it.
What you build
- Identify mode: point the camera, hear the denomination and feel a vibration pattern.
- Count mode: show notes one at a time and hear a running total.
- Change check: say or type the expected change, show the notes, hear whether it matches.
- Settings: speech rate, vibration only mode, language.
No login, no network, no data collection.
The technically interesting part
Data that looks like real use. You will build the dataset yourselves. Blind users do not frame photos: notes will be half in view, folded, upside down, under shop lighting or in front of a patterned cloth. Your training data must look like that. A negative class ("not a note") matters as much as the positive classes, because a confident wrong reading is worse than silence.
Decision rule. Speak only when the classifier is confident over several consecutive frames, and do not repeat the same note endlessly. Tune thresholds for low false readings, then report the trade-off between speed and error.
Accessibility engineering. Test every screen with TalkBack switched on, provide haptic feedback that works in noisy markets, and follow platform accessibility guidance rather than adding labels at the end.
How to evaluate it
- Top-1 accuracy per denomination on a held-out set photographed by people who did not collect the training set.
- False reading rate on non-note images and on folded or partly covered notes.
- Time to first spoken result, and model size, on a low-end Android phone.
- Task success and time with at least 5 blind or low-vision users, plus their comments, reported honestly.
Suggested plan
- Weeks 1 to 3: contact a visually impaired persons organisation, agree test sessions, list real needs.
- Weeks 4 to 9: collect and label images; train a small model; first accuracy numbers.
- Weeks 10 to 14: camera app, confidence rule, speech and vibration.
- Semester 2, weeks 1 to 5: counting and change modes, TalkBack pass.
- Weeks 6 to 10: user sessions and fixes.
- Weeks 11 to 14: evaluation and report.
Path to a product
Keep the app free for users. Funding can come from a bank or mobile wallet provider that wants an accessibility programme, or from an NGO grant, with the bank's name on the app. A second path is a small SDK other apps embed. Scale by adding currencies through a documented data collection and training recipe.
Scope warning
Do not attempt counterfeit detection; a phone camera cannot do it reliably and a wrong answer is harmful. Do not build a general scene describer. Do not add cloud features.
Learn what you need on Sefism
- Mobile roadmap : camera and accessibility on Android.
- Machine learning roadmap : dataset design, training and honest evaluation.
- Mobile cheat sheet : quick reference while building the camera screen.
- Android developers documentation : TalkBack and accessibility services.
- A11y Project checklist : a checklist to review every screen.
Get help with this idea
Getting access to real users early and planning the dataset are where teams usually need advice. A mentor can help you cut the scope to what your team can finish, choose between the stack options above, and design an evaluation your examiners will respect. You still build it; the help is on decisions, not code.
New here? Create a free account first.
Ideas and guidance, not finished projects
These are project ideas and scoping guidance, published free for students to use as a starting point. I do not build, write, or sell final-year projects, and I do not complete coursework for anyone. Take an idea, make it yours, and build it.
Not sure this one fits you?
I help students pick an idea that matches their skills and their deadline, then scope it down until it is finishable. Guidance only, never done for you.
Stuck on something specific?
Writing only gets you so far. If you want an answer to your situation rather than the general case, book a session and we will work through it together. Sessions are free for approved Sefism members, and a few slots open each week.
Follow along
New writing, resources and project ideas land here first.