New Research Helps Optimize Sensor Configuration and Sampling Rate for Accurate Movement Analysis

Human movement analysis is commonly based on data recorded with inertial measurement unit (IMU) sensors. The IMU data can be used for an algorithmic detection of different postures and movements. This serves for more detailed assessments of complex behaviors, such as daily activities.

Studies on human behavior in real-life environments need to find a balance between simplifying the recording settings and preserving sufficient analytic gains. Despite of wide use, it is poorly understood, what the trade-offs are between alternative recording configurations and the attainable analyses of naturalistic behavior.

The JMIR mHealth and uHealth recently published an article that sheds light on this topic that is of high interest to anyone using Movesense or other IMU sensors for classifying postures and movements.

In the article “Trade-Offs Between Simplifying Inertial Measurement Unit–Based Movement Recordings and the Attainability of Different Levels of Analyses: Systematic Assessment of Method Variations” Helsinki University Hospital researchers Manu Airaksinen, Okko Räsänen and Sampsa Vanhatalo assessed systematically the effects of different IMU recording configurations on the accuracy of posture and movement recognition.

The dataset used the study was recorded from spontaneously moving infants (N=41; age range 4–18 months). The recordings were made using a full-body wearable suit with four Movesense sensors attached, one at each limb.

The variables included the placement and number of IMU sensors, sampling frequency, and sensor modality. The analysis bechmark was human annotations of postures and movements. The reference IMU recording configuration included four IMU sensors that collected triaxial acceleration and gyroscope data at a frequency of 52 Hz. The study systematically tested how reducing the IMU data sampling rate, sensor modality, number of sensors, and different sensor placements affected the algorithmic classification of postures (N=7) and movements (N=9).

Low IMU sampling rate allows accurate human movement analysis

The results showed that reducing the number of sensors has a significant effect on classifier performance. For the postures and movements tracked in the study, single sensor configurations proved out non-feasible. Reducing sensor modalities to accelerometer only, that is, dropping gyroscope data, leads to a modest reduction in movement classification performance. However, the sampling frequency could be reduced from 52 to 6 Hz with negligible effects on the classifications.

The key findings are presented in the featured image of this post, originally published in JMIR mHealth and uHealth on Jun 3, 2025.

The findings of the study highlight the significant trade-offs between IMU recording configurations and the attainability of sufficiently reliable analyses at different levels. Notably, the single-sensor recordings employed in most of the literature and wearable solutions are of very limited use when assessing the key aspects of real-world movement behavior at relevant temporal resolutions.

The minimal configuration with an acceptable classifier performance of complex movements includes at least a combination of one upper and one lower extremity sensor, at least 13 Hz sampling frequency, and at least an accelerometer, but preferably also a gyroscope. These findings have direct implications for the design of future studies and wearable solutions that aim to quantify spontaneously occurring postures and movements in natural behaviors.

Implications for projects with Movesense sensors

The research findings provide excellent best practice guidance for using Movesense sensors in movement classification. The natural rules for reaching a good balance between accuracy and efficiency are

• Use a sensor configuration that provides sufficient data to reliably differentiate each movement class you are interested in.
• To simplify the measurement setup and to optimize the sensor power management and data analysis, use the lowest sample rate and set of sensor modalities that provides workable data.

Movesense sensors are often used for monitoring and classifying daily activity of adults in free-living space. For tracking general activity, energy expenditure, step count, and classifying certain distinct activities such as standing, walking, and running, a single sensor collecting acceleration data placed near the body’s center of mass generally provides acceptable detection accuracy.

However, as soon as the classification includes specific movements that include characteristic arm or leg motion, such as different gym exercises, running technique analysis, and physiotherapeutic exercises, a more specific sensor configuration is necessary.

In these situations, it is obvious that a sensor must be placed close to each part of the body whose movement is to be measured. If the monitored movement includes rotational elements, gyroscope data is needed.

Muscle-induced movement rarely includes frequency components above 20Hz, meaning that 52Hz sample rate is almost always sufficient, but often 26Hz or 13Hz as presented in this study are sufficient. However, if the movement includes impacts such as heel strike in running or contacts between athletes in team sports, a higher sample rate may be needed.

Read more:

Full article of Airaksinen et al. in JMIR mHealth and uHealth

A general flow of building a human movement analysis model with Movesense sensors and modern machine learning tools.