During my thesis my goal was to create a complex application that would be useful with a smartphone to determine parking and parking search, and to isolate them from other forms of road traffic, thus enabling them to use different applications to help them in the driver's parking process.
My thesis aims to solve the above problem by creating a learning algorithm based on a multi-layered perceptron, that is capable of making independent decisions in real time after the processing of a suitable training data set. Following the training process of my machine-learning algorithm, I present its accuracy, first by using the cross validation method on previously collected data and subjective accuracy, with real-time testing. During the presentation, I look at the general methods and the typical mistakes of creating and teaching neural networks and then introducing my own solution with the help of Weka JAVA class library, which provides a wide range of machine learning algorithms.
I have developed my own Android-based mobile application for the data collection and testing tasks needed to prepare my thesis, the essentials of which are presented in my dissertation The collection of motion data requires discussion of the descriptive capacity of the sensor types, their selection criteria, and their processing of data, as set out in the fourth chapter.
For the required data sampling and testing tasks I developed an Android based mobile application and I will introduce the related main concepts in this paper. The sampling of the moving data is the demonstration of the descriptive ability related to sensor types ,the viewpoint of their selection and also the processing of their data series with the help of sensor fusion.
Finally, I briefly describe the data collection measurements made in the semester and the main steps for the creation of the data collection and algorithm testing Android application.