Page 176 - AI for Good Innovate for Impact
P. 176

AI for Good Innovate for Impact



                      2�3     Future Work

                      Future works are anticipated to

                      1)   Build AI models for mimicry exercise recognition. There are already existing models for
                           detecting facial paralysis. These models can be combined as per our architecture. To
                           build these models, we need to collect a dataset. A dataset comprising predominantly
                           typical individuals along with a number of samples of individuals with facial paralysis
                           should be adequate for the task of training a model for mimicry recognition.
                      2)   A smart watch with the necessary hardware for collecting the video clips should be built.
                           Existing smart watch technology can be leveraged for building the hand movement
                           recognition system.


                      3�     Use Case Requirements

                      REQ-01: It is critical that the smartwatch be able to collect video of the face

                      REQ-02: It is critical that the video collected should have depth and/or colour information as
                      required by the AI model

                      REQ-03: It is critical that the smartwatch have a camera or other such hardware instrument that
                      points at the face to capture the video

                      REQ-04: It is expected that the smartwatch operate like a regular smartwatch that doesn’t
                      have an always-on display, i.e., it should detect and respond to the hand movement pointing
                      it towards their face by turning on the screen

                      REQ-05: It is critical that a video clip of the face is captured every time the user looks at the
                      watch

                      REQ-06: It is critical that captured video clips should be classified by AI as a normal face clip
                      or the wearer performing any of the mimicry exercises for face paralysis detection

                      REQ-07: It is critical that the mimicry exercises performed by the wearer are correctly classified

                      REQ-08: It is expected that Mimicry exercises are detected even when the user has not explicitly
                      interacted with the watch software to give an indication that they are going to do their exercise

                      REQ-09: It is critical that the sequence of mimicry exercises be analyzed by the AI for face
                      paralysis

                      REQ-10: It is expected that the software should keep track of the history of mimicry exercises
                      performed by the user

                      REQ-11: It is expected that the software prompts the user to do specific mimicry exercises if
                      they have not performed those in a while

















                  140
   171   172   173   174   175   176   177   178   179   180   181