There's hardly an election cycle in Nigeria without reports of blatant attempts of voter fraud. A very common example is under-aged individuals queuing up to vote. Such cases are usually resolved by election officials however, it takes a lot of time and resources to get to that point especially if voting centres are busy.
What if we could use image recognition technology to "scan" people on election day at voting centres to identify potential ineligible voters very rapidly? This will help reduce the amount of human effort to perform these checks which will lead to a smoother and faster voting process.
Using AI and facial recognition, we are able to detect all sorts of features from people's photos including: gender, age and even mood. If we can combine this with image data already present on voter registration records, it should be possible to automate verification and greatly speed up the job of identifying fraudulent voters.
What it does
- Voter verification - Cross check image taken by our app on polling day with image in PVC database. If it doesn't match, instantly flag
- Voter eligibility analysis - ** Age analysis - Based on image analysis, does the voter above the legal age? If not, flag as possible fraud
How we built it
- Web based app, built with a Node.js backend
- A "fake" INEC voters register, stored on Google Cloud Platform No-SQL Datastore
- Face API , an AI cognitive service by Microsoft Azure
- Accuracy - Can we rely on AI to reliably identify faces?
- Process - How can this tech be adapted into their existing election process?
- Hardware - Will the scanning device be a fixed device or handheld?
- Costs - How much more (or less) expensive will this tech make running elections?
- Privacy/Ethics - Will data be safely and reliably stored/transferred? Is it too intrusive?
What we learned
- Expand the tech beyond just accreditation to actual voting
- Use the principle of a central datastore for other government verification needs eg. taxes