From the Laboratory to the Field: IMU-Based Shot and Pass Detection in Football Training and Game Scenarios Using Deep Learning.

The applicability of sensor-based human activity recognition in sports has been repeatedly shown for laboratory settings. However, the transferability to real-world scenarios cannot be granted due to limitations on data and evaluation methods. On the example of football shot and pass detection against a null class we explore the influence of those factors for real-world event classification in field sports. For this purpose we compare the performance of an established Support Vector Machine (SVM) for laboratory settings from literature to the performance in three evaluation scenarios gradually evolving from laboratory settings to real-world scenarios. In addition, three different types of neural networks, namely a convolutional neural net (CNN), a long short term memory net (LSTM) and a convolutional LSTM (convLSTM) are compared. Results indicate that the SVM is not able to reliably solve the investigated three-class problem. In contrast, all deep learning models reach high classification scores showing the general feasibility of event detection in real-world sports scenarios using deep learning. The maximum performance with a weighted f1-score of 0.93 was reported by the CNN. The study provides valuable insights for sports assessment under practically relevant conditions. In particular, it shows that (1) the discriminative power of established features needs to be reevaluated when real-world conditions are assessed, (2) the selection of an appropriate dataset and evaluation method are both required to evaluate real-world applicability and (3) deep learning-based methods yield promising results for real-world HAR in sports despite high variations in the execution of activities.

Stride segmentation during free walk movements using multi-dimensional subsequence dynamic time warping on inertial sensor data.

Changes in gait patterns provide important information about individuals' health. To perform sensor based gait analysis, it is crucial to develop methodologies to automatically segment single strides from continuous movement sequences. In this study we developed an algorithm based on time-invariant template matching to isolate strides from inertial sensor signals. Shoe-mounted gyroscopes and accelerometers were used to record gait data from 40 elderly controls, 15 patients with Parkinson's disease and 15 geriatric patients. Each stride was manually labeled from a straight 40 m walk test and from a video monitored free walk sequence. A multi-dimensional subsequence Dynamic Time Warping (msDTW) approach was used to search for patterns matching a pre-defined stride template constructed from 25 elderly controls. F-measure of 98% (recall 98%, precision 98%) for 40 m walk tests and of 97% (recall 97%, precision 97%) for free walk tests were obtained for the three groups. Compared to conventional peak detection methods up to 15% F-measure improvement was shown. The msDTW proved to be robust for segmenting strides from both standardized gait tests and free walks. This approach may serve as a platform for individualized stride segmentation during activities of daily living.

Movement Speed Estimation Based on Foot Acceleration Patterns.

Wearable sensors are important in today's athlete training ecosystems and also for the monitoring of therapeutic rehabilitation processes or even the diagnosis of diseases. In the future, wearables will be integrated directly into clothing and require dedicated, low-energy consuming algorithms that still maintain high accuracy. We developed a novel algorithm for the task of movement speed determination based on wearables that track only the acceleration of one foot. It consists of three algorithm blocks that perform step segmentation, step detection and speed estimation, all having linear computation complexity and able to work in real-time on state-of-the-art embedded microprocessors. Using a reference dataset collected from a motion capturing device for nine subjects and 795 steps in total, a parametric regression algorithm was trained and evaluated using a comprehensive leave-one-subject-out crossvalidation. It is able to estimate the movement speed with a mean relative error of 6.9 ± 5.5 %. Furthermore, we evaluated our approach on lightgate-based reference measurements using 12 subjects and different running movement styles. Here, our algorithm achieved a mean relative error of 16.5 ± 8.4 %. A final evaluation with realistic football-specific movements in a three-aside cage-based soccer game was done with a GPS-based reference measurement system, where the speed profile over a 30 minutes game of our method had a Pearson correlation of 0.85 to the GPS-based reference speed profile.

Hierarchical, multi-sensor based classification of daily life activities: comparison with state-of-the-art algorithms using a benchmark dataset.

Insufficient physical activity is the 4th leading risk factor for mortality. Methods for assessing the individual daily life activity (DLA) are of major interest in order to monitor the current health status and to provide feedback about the individual quality of life. The conventional assessment of DLAs with self-reports induces problems like reliability, validity, and sensitivity. The assessment of DLAs with small and light-weight wearable sensors (e.g. inertial measurement units) provides a reliable and objective method. State-of-the-art human physical activity classification systems differ in e.g. the number and kind of sensors, the performed activities, and the sampling rate. Hence, it is difficult to compare newly proposed classification algorithms to existing approaches in literature and no commonly used dataset exists. We generated a publicly available benchmark dataset for the classification of DLAs. Inertial data were recorded with four sensor nodes, each consisting of a triaxial accelerometer and a triaxial gyroscope, placed on wrist, hip, chest, and ankle. Further, we developed a novel, hierarchical, multi-sensor based classification system for the distinction of a large set of DLAs. Our hierarchical classification system reached an overall mean classification rate of 89.6% and was diligently compared to existing state-of-the-art algorithms using our benchmark dataset. For future research, the dataset can be used in the evaluation process of new classification algorithms and could speed up the process of getting the best performing and most appropriate DLA classification system.

Subsequence dynamic time warping as a method for robust step segmentation using gyroscope signals of daily life activities.

The segmentation of gait signals into single steps is an important basis for objective gait analysis. Only a precise detection of step beginning and end enables the computation of step parameters like step height, variability and duration. A special challenge for the application is the accurateness of such an algorithm when based on signals from daily live activities. In this study, gyroscopes were attached laterally to sport shoes to collect gait data. For the automated step segmentation, subsequence Dynamic Time Warping was used. 35 healthy controls and ten patients with Parkinson's disease performed a four times ten meter walk. Furthermore 4 subjects were recorded during different daily life activities. The algorithm enabled counting steps, detecting precisely step beginning and end and rejecting other movements. Results showed a recognition rate of steps during ten meter walk exercises of 97.7% and in daily life activities of 86.7%. The segmentation procedure can be used for gait analysis from daily life activities and can constitute the basis for computation of precise step parameters. The algorithm is applicable for long-term gait monitoring as well as for analyzing gait abnormalities.