Changes in gait are associated with an increased risk of falling and may indicate the presence of movement disorders in connection with a neurological disease or age-related weakness. Continuous monitoring, based on inertial measurement unit (IMU) sensor data can be used to estimate gait parameters that indicate changes in gait. The monitoring from a waist-level IMU sensor is applicable for the assessment of such data, as it can be easily worn either as specific sensor-integrated belts or being suitable as a smartphone application. However, there is also the challenge that data from a waist-level mounted sensor provides a weaker representation of gait compared to signals from sensors placed closer to the legs and feet as sources of motion. Our work examines how well gait parameters can be estimated from the data of a waist worn IMU sensor. The approach employs the processing of the IMUs triaxial accelerometer and gyroscope data to detect gait events and estimate spatio-temporal gait parameters. In addition to the gait events heel strike (HS) and toe off (TO), it estimates the gait parameters step length, cadence, speed, stance time, swing time, and stride time. The results are compared to data measured by a GAITRite ® system as a reference. The present study investigates and compares the accuracy of two different approaches for gait event detection: a rule-based method and a machine learning (ML) approach based on a sequence-to-sequence convolutional neural network (CNN) architecture. The efficacy of these approaches in accurately determining gait parameters is evaluated using a dataset consisting of 21,193 recorded steps from 70 subjects, who performed 4,590 walks, each with a distance of approximately 4 meters. The ML-based algorithm performs better than the rule-based achieving an accuracy of over 99% in detecting gait events and estimating step lengths with a mean error of-0.12±4.49 cm.

Data anonymization is an essential prerequisite that enables data sharing in a privacy-preserving manner. However, anonymization affects the quality of the data and thus might affect the performance of later conducted data analysis. In this work, we investigate the influence of different image-based subject anonymization methods on the performance of the three common computer vision (CV) tasks keypoint detection, instance segmentation, and face detection. We compare the anonymization methods on how much they affect the performance of the CV tasks, as well as their degree of anonymization. Furthermore, a re-identification attack scenario is applied. The study findings indicate a clear trade-off between increased privacy results in decreased CV performance. Furthermore, DeepPrivacy2 (DP2) shows best results on keeping the performance of face detection high, while leading to a low recognition rate. However, the retrieval rate of our attack scenario which measures the reidentification rate remains high for all anonymization methods, which contrasts with respective low level of recognition.