Development of a smart mirror-based personal assistant by using raspberry Pi

Tracy, Yeo Hooi Yinn (2020) Development of a smart mirror-based personal assistant by using raspberry Pi. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

	Text (24 Pages) Development of a smart mirror-based personal assistant by using raspberry Pi.pdf - Submitted Version Download (673kB)
	Text (Full Text) Development of a smart mirror-based personal assistant by using raspberry Pi.pdf - Submitted Version Restricted to Repository staff only Download (2MB)

Abstract

Smart mirror systems are among the key features in the home of Internet of Things (IoT). A smart mirror-based personal assistant is a two-way mirror which has two functions. The front side of the glass has a reflection feature like a normal glass, while the back side is embedded with an electronic display that displays several information such as meeting schedule, weather, time and news updates. The current issue of the ordinary mirror is that it only shows any object in front of the mirror. People waste a lot of time standing in front of the mirror for washing, makeup and wearing clothes. Therefore, a smart mirror that displays several information is necessary to encourage efficient time management. The aim of this project is to develop a smart mirror-based personal assistant by using Raspberry Pi. In this project, a special feature of voice command is implemented to allow the users interact with the smart mirror without touching the screen. The system consists of a microcontroller (Raspberry Pi), sensors (Pi-camera, microphone), speaker, two- way mirror and LCD monitor. The pi-camera is used to recognize the registered users before the users are authenticated to use the system. A set of test case is conducted to make sure all the widgets or features are available to meet the functional requirements. The result of the test case shows that the system able to display all the features at the specified layout on the LCD Monitor. The accuracy of the voice and face recognition is analyzed through a series of real testing. Word Error Rate is the calculation used to analyse the accuracy of voice recognition. The results shows that the Word Error Rate is 5% in noisy room while 1.67% in silent room, which are in reasonably accurate as the system misinterpret the input speech. The average response time of Google Assistant is 2.61 seconds which are relatively quick to perform real-time transcription. The accuracy of the face recognition for known and unknown faces are 90% and 80% respectively but the accuracy of the face recognition is affects by the illumination problem. This is based on the result that shows 90% of faces recognized by the system in a bright while only 30% of faces recognized in dark room. To conclude, the developed system can accurately recognize the users and perform the voice commands as required. This project has potential to be installed in hotel lobby, bathroom and living room as well as can be integrated with modern virtual or augmented reality system. Example application of virtual reality system is virtual fashion consultant to provide user to analyze and recommend the virtually try outfits, makeup and handbags in front of a mirror.

Actions (login required)

View Item