Context-sensitive smart glasses monitoring wear position and activity for therapy compliance - a proof of concept

Purpose To improve the acceptance and compliance of treatment of amblyopia, the aim of this study was to show that it is feasible to design an electronic frame for context-sensitive liquid crystal glasses, which can measure the state of wear position in a robust manner and detect distinct motion patterns for activity recognition. Methods Different temple designs with integrated temperature and capacitive sensors were developed to realize the detection of the state of wear position to distinguish three states (correct position/wrong position/glasses taken off). The electronic glasses frame was further designed as a tool for accelerometer data acquisition, which was used for algorithm development for activity classification. For this purpose, training data of 20 voluntary healthy adult subjects (5 females, 15 males) were recorded and a 10-fold cross-validation was computed for classifier selection. In order to perform functional testing of the electronic glasses frame, a proof of concept study was performed in a small group of healthy adults. Four healthy adult subjects (2 females, 2 males) were included to wear the electronic glasses frame and to protocol their activities in their everyday life according to a defined test protocol. Individual and averaged results for the precision of the state of wear position detection and of the activity recognition were calculated. Results Context-sensitive control algorithms were developed which detected the state of wear position and activity in a proof of concept. The pilot study revealed an average of 91.4% agreement of the detected states of wear position. The activity recognition match was 82.2% when applying an additional filter criterion. Removing the glasses was always detected 100% correctly. Conclusion The principles investigated are suitable for detecting the glasses' state of wear position and for recognizing the wearer´s activity in a smart glasses concept.