Design and Evaluation of Smart Glasses for Food Intake and Physical Activity Classification

The overall goal of this procedure is to design and manufacture a glasses type wearable device that detects the patterns of temporalis muscle activities during food intake and other physical activities. This method can help answer key questions in the field of monitoring of ingestible behaviors such food intake detection during a meal. The main advantage of this technique is that the patterns of temporalis muscle activity during the food intake can be detected automatically and objectively using a pair of glasses.

The manufacturing procedure will be demonstrated by Wonjoon Oh, a mechanical engineer at our company. The data collection procedure will be done by Hak Yung Gu, who is also at our company. Key to the experiment are the circuit modules which must be manufactured.

There are two. One for the left and right temples in a pair of glasses. The left module contains a load cell, connector, instrumentation amplifier, voltage regulator, and battery connector.

The right module has a micro-controller with WiFi, UART connector, multiplexer, amplifier, connector, and load cell. In addition to the circuits, glasses frames have to be fabricated. The frames must be able to support the printed circuit board modules.

To create the frames, use a 3-D modeling tool to draw a model of the headpiece of the glasses and it's two temple pieces. Export the drawing to CAD's stereo lithography files for 3-D printing. When printing the frame parts use a carbon filament with a 250 degree celsius nozzle temperature and an 80 degree celsius bed temperature.

After printing, work to bend the temples inward. Use a hot air blower set at 180 degrees celsius to heat a piece. Once heated, bend the temple inward about 15 degrees.

Bend each temple piece inward before continuing. Gather all the pieces to assemble the glasses. These include left and right printed circuit boards, left and right temples, and a headpiece.

Begin with the right temple components. Insert the right circuit board in the right temple. Secure it in place with a bolt.

Secure the left circuit board and left temple in the same way. Next, attach the headpiece and one of the temples at the appropriate hinge joint and secure it with a bolt. Continue by attaching the other temple to the headpiece.

When done, connect the left and right circuits using 3 pin connecting wires. For this design, connect a battery to the left circuit. Secure it with adhesive tape.

Complete the glasses by applying rubber tape at the tip and on the nose pads to increase friction for the wearer. At this point, prepare for data acquisition. At a computer, run the GlasSense_Server project and connect the frame circuit module via USB.

Flash the GlasSense_Server software into the micro-controller. This project uses a smartphone to receive the data. Run the GlasSense_Client software on the computer and connect the smartphone via a USB connector.

Build the data receiving application on the smartphone. For the study, identify a subject and select suitably sized glasses. Have the subject try on the glasses.

If necessary, use the support bolts on the frame to fine tune the fit. The next step is to have the subject collect data for different activities. For the sedentary rest activity, have the subject sit in a chair to read with the glasses on.

Record the activity using the smartphone application for 120 seconds. Allow head movement but avoid movement of the entire body. After 120 seconds, stop the recording and take off the glasses for a 60 second break.

Repeat recording data and taking a break four times. Follow a similar procedure for the walking activity by having the subject stand on treadmill and wear the glasses. Start the treadmill at 4.5 km per hour and have the subject record for 120 seconds.

When done take a 60 second break with the glasses off. There are also sedentary chewing and chewing while walking activities. For these prepare toasted bread in 20x20 mm squares and have chewing jellies ready.

Arrange for the subject to sit or be on the treadmill, as appropriate, with the food nearby. While wearing the glasses, start the activity and data recording. Have the subject eat bread while chewing normally.

After 120 seconds, take a 60 second break with the glasses off. Next, perform a sedentary talking activity. For this, collect data as the subject sits and reads a book out loud in a normal tone and speed.

In the sedentary wink activity, arrange for the subject to wink in response to a bell set to go off every three seconds. These data are from one subject performing each activity. The horizontal axis is the time in seconds.

Each vertical axis is the measured force after the median of the ten measurements is subtracted. The blue curves are for the left side and the red curves are for the right side of the face. As might be expected, the maximum amplitudes of the chewing activities are large when compared to the sedentary rest, sedentary talking, and walking activities.

Data of this type are transformed into input to a machine learning classification technique. After each development, this technique can be potentially useful for automatic and objective monitoring of ingestible behaviors with higher accuracy as a practical means to treat ingestive problems.