Contribution
Our solution reveals behavior of visitors in retail and helps to adjust the business right to the their needs. Faces provide real-time statistics with fully anonymized information about the customers. The system tracks their movement, detects duration of stay and predicts their basic demographic information such as gender and age. No worry about privacy policy, video data are not stored not sent anywhere.
Compared to other similar applications, Faces device does not need a dedicated server with expensive GPU cards. It only needs Raspberry Pi – which achieves a competitive performance. Faces detects people up to nine meters far away from the camera.
An accuracy of gender prediction is 95 percent and an average error of age prediction is less than seven years.
Since it was trained on our specifically annotated data set, it works even in sub-optimal conditions such as bad light or fast moving people.