Posture Monitoring and Wheelchair Control Using Optical Fiber Pressure Sensors
Name: AURA XIMENA GONZALEZ CELY
Publication date: 02/03/2023
Advisor:
Name |
Role |
|---|---|
| CAMILO ARTURO RODRIGUEZ DIAZ | Co-advisor * |
| TEODIANO FREIRE BASTOS FILHO | Advisor * |
Examining board:
| Name |
Role |
|---|---|
| CAMILO ARTURO RODRIGUEZ DIAZ | Co advisor * |
| RICARDO CARMINATI DE MELLO | Internal Examiner * |
| TEODIANO FREIRE BASTOS FILHO | Advisor * |
Summary: A posture monitoring system and wheelchair control based on Polymer Optical Fiber (POF) pressure sensors was developed and installed on an electric-powered wheelchair and a neck pillow. The static characteristics of the POF-based pressure sensor are a response time of 33 µs, a mean linearity of 99.11%, a mean resolution of 5.95 mV, and a sensitivity of 3.9 mV/Kg. A characterization per sensor, line, and matrix of sensors resulted in a linear response, but with variations in the offset of the sensor because of the fiber delays returning to its original state. Moreover, a posture monitoring system was developed both in an offline and online stage using Machine Learning (ML) techniques. The offline stage used a Butterworth low-pass filter and a Common Average Reference (CAR) filter in the pre-processing stage. The best result was obtained with the k-Nearest Neighbors (kNN) algorithm, with an accuracy of 99.16% with one-person data. After that, ten healthy people participated in the dataset construction, obtaining a classifiers accuracy of 98.49% using the Extra Tree Classifier (ETC) algorithm with an execution time of 2 s. The online classification used a kNN model obtaining an accuracy of 96.87 % with a mean prediction time of 117 ms. Moreover, neck control was implemented in offline and online stages. A features comparison was conducted in the offline stage, and the best accuracy was 85.50% with a Decision Tree (DT) algorithm. In the online stage, a Direct Current (DC) filter was implemented due to the natural response of the POF. A fuzzy logic controller was developed and validated, obtaining an execution time of 26 ms, moving in four directions (forward, right, left, and stop) when the user was out of the wheelchair. In addition, a head fuzzy controller using a motion capture system was implemented with an execution time of 24 ms moving in five directions including backward. The user was sitting in a wheelchair and the validation of the controller was made in a laboratory room. Future works will be focused on the improvement of the sensors, the inclusion of an obstacle avoidance system, and the neck controller
enhancement.
