Vision-Based Feedback on Correct Sensor Placement in Medical Studies

Wearable sensors simplify complex tasks involving continuous monitoring of physiological and environmental data or user interaction with smart systems. Automatic recognition of correct placement of these sensors is essential in situations where non-expert users are employing them. However, this is a very challenging problem because of the different environment conditions, heterogeneous people features and sensor placement characteristics. In this paper we address this problem for a neuroheadband sensor, a device designed for recording electroencephalogram (EEG) data, which can be used in Brain-Robot interfaces or different medical studies. Our work presents a vision-based system that provides feedback to assist users to correctly position the neuroheadband. The system employs computer vision techniques to analyze facial landmarks and headband location. We evaluate two significant aspects of the prototype presented: (1) the predictions and feedback accuracy, and (2) the feasibility of running the approach in computationally constrained environments. The experimental results demonstrate that our vision system correctly detects placement errors in 94.29% of the test cases and present execution times compatible with the embedded devices where this feedback system should run, such as standard low-end tablets or laptops.