| Abstract [eng] |
This work analyses various dangerous maneuvers that are performed with micro-mobility vehicles, which can harm the user or pedestrians. A system is developed, which is capable of collecting data from various sensors, several methods are evaluated, which are used in detecting maneuvers from sensor data, identifying them and determining their danger level. The aim of this work – conduct research on the most common micro-mobility dangerous maneuvers, identify them, with the help of sensors record them and using selected algorithms, identify, classify, and compare them. The algorithms should require minimal computational resources, be suitable for signal recognition, and perform calculations as quickly as possible. The first part analyzes the most common sensors used in mobility tasks, describing their principles of operation and measured quantities. Literature sources and related research on object motion and maneuver recognition are also analyzed. Finally, an overview of a company which produces smart scooters with integrated safety features is provided. The second part discusses two data recording methods that were used: a smartphone and developed embedded system. First, the mobile phone set up is described for collecting accelerometer data, highlighting its pros and cons, followed by data visualization and preliminary analysis. Second, the development process of the second data acquiring system is provided, which solves the problems of the first system. The detailed structure and components of the second experiment stand are described along with the algorithm of entire system. Later, dangerous maneuvers are identified, their characteristics described, and experiments are conducted to determine the danger level of each maneuver, a database is created. The sequence of feature extraction from experiment data is described, filtering methods are compared, and the optimal filtering method is selected. Additionally, the standard deviation of each maneuver feature is calculated to determine its overall quality. Finally, five algorithms used for maneuver detection are described. In the third part, the results, obtained using “Matlab” software package, are analyzed. The examined methods are compared in case of correct and incorrect maneuver detection. From the obtained results, the algorithm or method that showed the best results for each researched dangerous maneuver is selected. |
