Results
The Facial Landmark recognition accurately determined eye state, closed or open, and identified the direction the head was positioned.
Similarly, mobile device positioning was able to accurately determine when a phone was raised in order to be used for conversation. However, in the training data, the first 3 elements are always the accelerator changes of the beginning of the movement. While in the practice, that is not always the case as we are collecting data continuously. To improve the performance, we set a rather small prediction score threshold to improve the recall of the model. During our simple test, about 70% of phone raise actions are detected.
These findings were able to be combined and, when appropriate, notified the end-user.
Similarly, mobile device positioning was able to accurately determine when a phone was raised in order to be used for conversation. However, in the training data, the first 3 elements are always the accelerator changes of the beginning of the movement. While in the practice, that is not always the case as we are collecting data continuously. To improve the performance, we set a rather small prediction score threshold to improve the recall of the model. During our simple test, about 70% of phone raise actions are detected.
These findings were able to be combined and, when appropriate, notified the end-user.