There is a raising demand for player statistics in the world of football. With the developments over the last years in wearable sensors, Human Activity Recognition (HAR) based on wearable IMU sensors can be used to tackle this problem. This thesis builds upon an earlier research done for this topic, where an end-to-end pipeline based on deep learning was created that is able to be trained and used for football activity recognition. The goal is to test and improve said pipeline. This was done by adding change of directions (COD’s) to the classifiable activities. Furthermore, run velocities were build as a spectrum with several categories depending on the speed. A combination of convolutional and recurrent layers resulted in test accuracies up to 88.9%. Afterwards, the pipeline was used to evaluate larger datasets containing football drill and a football physiotherapy training. For this a sliding window evaluation procedure was proposed. These evaluations gave promising results. Many actions and football related activities could be recognized, however many smaller, shorter actions were missed. This can be seen as lack in trainingdata. In this data, little activities with the ball were present. Hence the deep learning models could not be trained accordingly. Later, it was researched if additional training of activities with ball increased the evaluation. This was indeed confirmed, since the evaluations showed more detailed and realistic results. Including even more additional trainingdata, could result in the pipeline performing reliably in real-life football scenario’s.

been given to Human Activity Recognition (HAR) based on signals obtained by IMUs placed on different body parts. This thesis studies the usage of Deep Learning-based models to recognize different football activities in an accurate, robust, and fast manner. Several deep architectures were trained with data captured with IMU sensors placed on football players' bodies and their performances were compared. A combination of convolutional layers followed by recurrent (bidirectional) LSTM layers showed to achieve the best results with up to 96.71% of accuracy. When using normalized data via a calibration recording, these accuracies increased up to 98.2%. Results showed that deep learning models performed better in evaluation time and prediction accuracy than traditional machine learning algorithms. An end-to-end pipeline for football activity recognition was developed that can be extended to any other HAR task. With it, not only the training was explored, but also a sliding window evaluation procedure was proposed that can be used to efficiently analyze unseen IMU recordings and recognize the activities there present. This pipeline showed to be fast and robust especially with signals consistently calibrated and can be used to recognize movements on a real-time basis. It can be concluded that a combination of deep learning models and a sliding window evaluation procedure is suitable for fast and accurate HAR tasks and can be used as input for research on injury prevention. By training and evaluating the models with more data, ideally from uncontrolled experiments such as football matches, we expect to further improve the generalization property of the classifiers.