|
|
Utilization of Motion Sensor Data for User Activity Analysis
Eršek, Martin ; Zemčík, Pavel (referee) ; Beran, Vítězslav (advisor)
This bachelor thesis aims to provide a design and implementation of an algorithm for analysis of user activity based on data from motion sensors. The thesis explores possibilities of classification and counting repetitions of 7 basic body-weight exercises, namely: push ups, squats, planks, sit-ups, seated knee raises, tricep dips and lunges. Data from motion sensors are collected by a mobile device located in top pocket of exercising user's trousers. Selection of used methods and their parameters as well as number and type of extracted features is chosen with regard to low computational complexity. When designing a solution, emphasis was put on the fact that it is irrelevant how the device is positioned in the user's pocket. For the thesis, a dataset containing 7 training sessions from 4 different users was created. Designed method was implemented as a desktop application with Command Line Interface and consequently validated on the created dataset. The solution was able to reach metrics of F1-score in range 45.3 % - 74.9 % for analysis and counting repetitions of an unseen user's training session. For an unseen training session of a known user, the metrics of F1-score was up to 94 %.
|
|
|
Utilization of Motion Sensor Data for User Activity Analysis
Eršek, Martin ; Zemčík, Pavel (referee) ; Beran, Vítězslav (advisor)
This bachelor thesis aims to provide a design and implementation of an algorithm for analysis of user activity based on data from motion sensors. The thesis explores possibilities of classification and counting repetitions of 7 basic body-weight exercises, namely: push ups, squats, planks, sit-ups, seated knee raises, tricep dips and lunges. Data from motion sensors are collected by a mobile device located in top pocket of exercising user's trousers. Selection of used methods and their parameters as well as number and type of extracted features is chosen with regard to low computational complexity. When designing a solution, emphasis was put on the fact that it is irrelevant how the device is positioned in the user's pocket. For the thesis, a dataset containing 7 training sessions from 4 different users was created. Designed method was implemented as a desktop application with Command Line Interface and consequently validated on the created dataset. The solution was able to reach metrics of F1-score in range 45.3 % - 74.9 % for analysis and counting repetitions of an unseen user's training session. For an unseen training session of a known user, the metrics of F1-score was up to 94 %.
|
guest ::
