Improvement of three dimensional path finding systems using inertia based measurement units

Inertial measurement units (IMU) are used to determine position, speed, or orientation in a wide range of applications such as robotics, consumer electronics, and healthcare. IMUs can offer precise measurements by combining the information obtained from the accelerometer, gyroscope, and magnetometer sensors. The major disadvantage of IMUs is that they always measure changes relative to themselves and are prone to noise-induced error. This thesis aims to improve the acceleration data of IMU sensors using some filters and machine learning approaches. The improvement is carried out in two stages. In the first stage, some filters are applied to the data to decrease the noise. In the second stage, the data is marked and an improvement model is implemented using the machine learning approach. A high precision industrial robot from KUKA is used to create the data set. A mounting apparatus is designed to fix the IMU sensors and necessary electronic cards to the industrial robot flange. In the study, the acceleration data of the endpoint of the industrial robot is used as the reference acceleration value. A program is created in the Python programming language using a multiprocessing approach to synchronously collect the acceleration data measured by the sensors and the reference acceleration data from the robot controller. As the first filtering, the Kalman filter has been applied to each IMU sensor's output separately. Then, a median filter and a conditional mean filter is applied consecutively to the output of Kalman. The data set is obtained by matching this filtered data with the reference acceleration data from the industrial robot. Three different DNN regression models are implemented and tested for these data sets. According to the comparative results, the applied filtering and regression model improved the acceleration data.