Development of the gesture recognition systems requires a dataset that contains a gesture vocabulary we want to recognize. However, the previous research results has not been published the created data sets, thus the developers had to create their database. Building a large gesture database takes up a lot of work and time, therefore researchers create small datasets only. Without a public database, it is not possible to compare the research results.
This paper aims to build a benchmark database, which can be used the development of gesture recognition systems, and it can be use for various methods of gesture recognition, and it allows an objective comparison for the performance of the developed systems.
In this document, I present the published accelerometer based gesture recognition systems, procedures and their applications, and the created datasets for these projects. I determined a large gesture vocabulary for the data collection task. Furthermore, I describe the technical parameters of the accelerometer mounted devices, which are important in the gesture recognition.
To store gesture samples in a database, I defined a datastructure, that allows using data for different types of procedures of the gesture recognition at the same time. To effectively build a large gesture database, I developed an application that allows collecting adequate number of gestures within a short time, ensures the automated recording of gestures and prevents the erroneous recording of data. Finally, I present the application and I set the collected gesture samples against the published dataset. The established database contains 13800 samples from 30 participants, with this size, this is the largest published database.