# Designing a Gait Recognition Algorithm for Older Adults Using Mobility Aids: Prospective Cohort Study

## TL;DR

This study developed a gait recognition algorithm tailored for older adults, including those using mobility aids, using wrist-worn devices to improve accuracy and usability.

## Contribution

A novel gait recognition algorithm calibrated for older adults using mobility aids, leveraging harmonic patterns and accelerometer data.

## Key findings

- The algorithm successfully detected walking patterns in older adults with and without mobility aids.
- Adjusting parameters like signal amplitude and frequency range improved detection accuracy for older adults.
- The algorithm is computationally lightweight and relies solely on accelerometer data.

## Abstract

Maintaining mobility is important for older adults to retain independence and reduce fall risk. Wearable technology such as fitness trackers and smartwatches can track physical activity. Unfortunately, gait recognition algorithms are often calibrated using younger adults and are not accurate for older adults, especially when using mobility aids.

Our goal was to develop a gait recognition algorithm capable of detecting the walking patterns of older adults that is robust to using mobility aids. Wrist-worn wearable devices were used to maximize the ubiquity of the approach.

We collected walking and other daily activity data on 17 independent older adults to develop a gait recognition algorithm. Ten participants used mobility aids (ie, 5 cane users, 4 rollator users, 1 walker user). We calibrated a heuristic-based “one-size-fits-most” algorithm leveraging the harmonic patterns associated with walking to recognize the walking patterns of our cohort. This algorithm is computationally lightweight and relies only on accelerometer data. We used hyperparameter tuning using a Parzen tree estimator to find the optimal parameters in a leave-one-subject-out fashion.

The calibration process was required for this algorithm to detect walking. The signal amplitude threshold lowered from 0.3g to 0.1g to detect the more subtle walking patterns of older adults. The walking frequency range widened from [1.4Hz, 2.3Hz] to [0.8Hz, 2.8Hz], showing that older adults walk more slowly. The ratio for superharmonics increased from 1.4 to 77.11. Analyzing the false positive rate for the other daily activity classes implies that these superharmonics are artifacts of back-and-forth arm motions that characterize walking in our collected data. Additionally, we report the performance metrics of sensitivity, specificity, and F1-score to evaluate our algorithm. Sensitivity increased from 0.11 to 0.73; F1-score increased from 0.15 to 0.73; and specificity decreased from 0.99 to 0.75.

This study successfully recognized the walking patterns of older adults with or without mobility aids. The performance metrics show that this algorithm has promise for being used to monitor physical activity. This approach is computationally lightweight and explainable. Our calibration approach can be adopted to tune to new populations and has a low barrier to entry for adopting a new technology due to the sole reliance on accelerometer data which is a standard sensor in wearable devices. The most noteworthy parameter adjustment is the ratio for superharmonics. Low values for the subharmonic and superharmonic ratios cause the algorithm not to detect walking in our older adult data. We validated the algorithm on ten mobility aid users. A larger study with more participants using mobility aids is necessary to conduct a deeper analysis on what parameters work best for this population. Future work includes validating the algorithm’s ability to estimate step counts and measure physical activity in real-world settings.