Visual Person Identification Device using Raspberry Pi

People with low vision find it difficult to socialize publicly due to lack of face recognition ability. Robust assistive solutions are available but either those are too heavy on the pocket or are not wearable. In this paper, we have proposed a low cost lightweight wearable device, based on raspberry pi 3 that uses enhanced machine learning algorithms. The device has two user-selectable operation modes; recognition mode and new entry or training mode. In recognition mode, a combination of Haar cascade and Hog features are used for face detection and recognition. The accuracy of recognition system is further improved by capturing a burst of five images for each subject. Face detection and recognition algorithms are next applied to each frame and the name with maximum votes is announced on the earphones. In case of unrecognized faces, system announces Unknown. For the training mode, a web based user interface has been proposed that can help the user for enrolling and updating entries, into the dataset. The system has been tested in real time environment. Results show the accuracy of the proposed methodology.