09 / 2023 – 02 / 2024
Wearable technologies and sensors have become viable sources of information on users’ health conditions . Among various sensors, footworn IMUs, due to their low cost, ease of use, and performance are considered a reliable option for measuring activities, specifically in the gait analysis domain [2, 3, 4]. Accordingly, footworn IMUs can provide accessible and continuous information about the status and progress of movement-related disorders such as Parkinson’s disease (PD) . The objective long-term monitoring provided by IMUs can assist clinicians with diagnoses and treatment procedures [3, 6]. The state-of-the-art home monitoring systems, using foot-worn IMUs, record the entire wear time and apply an offline gait analysis pipeline to the data after the recording has ended. These systems have two main disadvantages. First, the recorded sessions include irrelevant data that contain most of the recording duration. A motion analysis on daily home monitoring recordings of 28 patients with PD, from the FallRiskPD dataset, shows that 69.8% of the data are non-motion, where the signals are stationary. 20.2% of the data is related to any motion, including non-gait motion, single strides, and bouts less than four strides. Only 10.0% of the data contain bouts more than 4 strides. The constant recording of all sensors result in an extreme and inessential computational cost in parameter calculation, as well as memory and battery consumption. Second, they lack the ability to provide information or feedback to patients or their caregivers. Real-time feedback can assist patients in the occurrence of an event, such as a fall or a freeze. A real-time gait analysis system can address the two mentioned disadvantages. It can identify, store, and process only the relevant gait data. This decreases the size of the stored signals and therefore, reduces the transmission duration and the run time of the pipeline. Consequently, the power consumption of the sensor system will be reduced which increases their recording run time. Moreover, it enables measuring the patient’s gait performance and conducting limited interventions in case of certain events such as cueing the patient during a freezing episode. In our literature research, focused on real-time gait analysis utilizing Inertial Measurement Units (IMUs) at the lower extremities, we encountered a diverse array of publications addressing this intriguing domain. While a substantial number of studies have broached the topic, it’s noteworthy that only a fraction of these studies have implemented their methodologies onto microcontrollers, signifying a gap between theory and practical implementation [7, 8, 9, 10, 11, 12]. Conversely, many have opted to transfer the data to a computer for subsequent analysis. To the best of our knowledge, the majority of these studies measured events such as toe-off and heel-strike without any stride segmentation algorithm. Moreover, they validated the performance of their algorithms only on the data recorded in laboratory settings. These limitations invariably hamper the depth of their analyses, restricting insights solely to detected events within controlled environments. Only a handful of the publications address hardware characteristics important for home monitoring applications, such as run-time, computational complexity, memory consumption, and battery usage. Even fewer have implemented motion detection alongside their real-time gait analysis pipeline. Therefore, their systems constantly calculated the gait parameters, which is unsuitable for home monitoring applications. Wu et al. performed a run-time analysis of their algorithms and concluded that the system could only run at the maximum sampling rate of 33 Hz. While this rate holds merit within the context of gait analysis, its adequacy may vary for other applications such as event prediction in patients with PD . Based on our research, only Chang et al. calculated the gait parameters as part of the real-time system instead of only detecting the events. However, they also tested their system in a small cohort and under controlled conditions . Also, the size and the placement of their sensor system makes it undesirable for long term recordings. Rui Hua et al. developed a gait detection and analysis pipeline. This allowed the system only to record the gait data to save energy and memory. Their method showed high efficiency regarding energy consumption and could run for several days. However, its performance is evaluated on lab data and a few participants. Furthermore, due to a deep sleep mode, it detects only 10.91% of the strides . This project highlights the relevance of an efficient, accurate gait analysis system applicable for real-time home monitoring. On the other hand, an adequate tradeoff between accuracy and computational complexity has yet to be found. This thesis aims to follow the two mentioned objectives while taking the gap in the literature into consideration.
1. Real-Time Motion Detection Algorithm: The proposed algorithm should not only be able to distinguish between motion and stationary periods but also detect gait bouts and prepare the module for a real-time gait parameter calculation. The data logging should be stopped during stationary periods to reduce memory demands. In the future, this algorithm can be utilized as a basis for a low-power mode, where the system’s functionality is reduced to a minimum during nonmotion periods to save energy. It is noteworthy that the physical implementation of a low-power mode (sleep mode) on the IMUs is out of the scope of this thesis.
2. Real-Time Gait Analysis: The system should be able to report the least computationally expensive gait parameters (mostly temporal parameters). The performance of the algorithms will be evaluated based on their accuracies and the delays posed by their complexities. The algorithms will be implemented on the ublox ANNA-B402 module integrable into wearable sensors. A subset of the FallRiskPD dataset will be used as a real-time simulation feed for algorithms’ development and measurements in unsupervised daily living motions . Afterwards, we implement the desired methods on a prototype sensor, integrated with the processor, as an actual real-time system. Finally, we evaluate real-time gait analysis during standardized test walks such as 4×10 meters. The obtained parameters will be compared to parameters calculated by gaitmap from the same test walk. A fine analysis of the comparison between the two systems will be carried out to determine the weaknesses and strengths of the system. This includes differences in gait parameters and the detection of events. It should be noted, that the development of the IMU integrated with the microcontroller is in progress. Therefore, the feasibility of a new data collection with the implemented algorithms on the final module (IMU integrated with the processor) depends on the availability of the hardware. Otherwise, we use the simulator to transfer data as real-time feeds to the processor to test the performance.
 Eduardo Teixeira et al. “Wearable Devices for Physical Activity and Healthcare Monitoring in Elderly People: A Critical Review”. In: Geriatrics (Basel, Switzerland) 6.2 (2021).
 Sebastijan Sprager and Matjaz B. Juric. “Inertial Sensor-Based Gait Recognition: A Review”. In: Sensors (Basel, Switzerland) 15.9 (2015), pp. 22089–22127.
 Pei-Hao Chen et al. “Gait Disorders in Parkinson’s Disease: Assessment and Management”. In: International Journal of Gerontology 7.4 (2013), pp. 189–193.
 Silvia Del Din et al. “Free-living monitoring of Parkinson’s disease: Lessons from the field”. In: Movement disorders : official journal of the Movement Disorder Society 31.9 (2016), pp. 1293–1313.
 Lazzaro Di Biase et al. “Gait Analysis in Parkinson’s Disease: An Overview of the Most Accurate Markers for Diagnosis and Symptoms Monitoring”. In: Sensors (Basel, Switzerland) 20.12 (2020).
 Shobha S. Rao, Laura A. Hofmann, and Amer Shakil. “Parkinson’s disease: diagnosis and treatment”. In: American family physician 74 (2006), pp. 2046–2054.
 H. F. Maqbool et al. “A Real-Time Gait Event Detection for Lower Limb Prosthesis Control and Evaluation”. In: IEEE transactions on neural systems and rehabilitation engineering : a publication of the IEEE Engineering in Medicine and Biology Society 25.9 (2017), pp. 1500–1509.
 I. P. Pappas et al. “A reliable gait phase detection system”. In: IEEE transactions on neural systems and rehabilitation engineering : a publication of the IEEE Engineering in Medicine and Biology Society 9.2 (2001), pp. 113–125
 Andrea Mannini, Vincenzo Genovese, and Angelo Maria Sabatini. “Online decoding of hidden Markov models for gait event detection using foot-mounted gyroscopes”. In: IEEE journal of biomedical and health informatics 18.4 (2014), pp. 1122–1130.
 Laurent Chiasson-Poirier et al. “Detecting Gait Events from Accelerations Using Reservoir Computing”. In: Sensors (Basel, Switzerland) 22.19 (2022).
 Joana Figueiredo et al. “Gait Event Detection in Controlled and Real-Life Situations: Repeated Measures From Healthy Subjects”. In: IEEE transactions on neural systems and rehabilitation engineering : a publication of the IEEE Engineering in Medicine and Biology Society 26.10 (2018), pp. 1945–1956.
 Jiaen Wu et al. “Real-Time Gait Phase Detection on Wearable Devices for Real-World Free-Living Gait”. In: IEEE journal of biomedical and health informatics PP (2022).
 Che-Wei Chang et al. “IMU-Based Real Time Four Type Gait Analysis and Classification and Circuit Implementation”. In: IEEE Sensors Journal (2022), pp. 1–4.
 Rui Hua and Ya Wang. “Monitoring Insole (MONI): A Low Power Solution Toward Daily Gait Monitoring and Analysis”. In: IEEE Sensors Journal 19.15 (2019), pp. 6410–6420.
 Martin Ullrich et al. “Fall Risk Prediction in Parkinson’s Disease Using Real-World Inertial Sensor Gait Data”. In: IEEE journal of biomedical and health informatics 27.1 (2023), pp. 319–328.
 Alexander Rampp et al. “Inertial sensor-based stride parameter calculation from gait sequences in geriatric patients”. In: IEEE transactions on bio-medical engineering 62.4 (2015), pp. 1089–1097.