Gesture recognition is a fundamental part of future human-machine interaction in industrial environments as well as in the Internet of Things. For the latter, gesture recognition has to be performed by wearable devices, which usually will come with very limited resources and relatively simple sensors. The low classification effort needed for an artificial neural network (ANN) once it is suitably trained makes this class of classifiers a promising candidate for gesture recognition on resource-limited platforms. In this paper we present a light-weight framework which enables gesture classification based on IMU data by means of an artificial neural network (ANN). We focus on low-complexity approaches for the data preprocessing and show that a considerable classification accuracy is achievable following these approaches. Additionally, we show that the application of well chosen augmentation methods on the training data can substantially increase the classification accuracy and significantly reduce the number of necessary neurons as well as the necessary amount of training data.