Method to Improve Gait Speed Assessment for Low Frame Rate AI Enabled Visual Sensor

The research potential of home-based autonomous health assessment has grown in recent times with the decline in caregivers for the aging population. There are several verified methods for automatic gait assessment using various kinds of sensors and cameras, however each of them comes with their own limitations. Previous gait speed assessments using an innovative privacy respecting visual sensor showed potential but were limited by the camera’s asynchronous and low frame rate. This paper extends this work with methods focused on reducing these limitations. This paper proposes a method to estimate the lost or dropped frames originally captured by the visual sensor through linear, quadratic, and cubic regression. Bisection methods are used on these regression polynomials to calculate the time taken for walking a predetermined distance, thence estimating the walking speed. The proposed method successfully regenerates the lost data back to the frame rate of 30 frames/sec whilst reducing the mean percentage error to ~6% and ~11% from ~13% for quadratic and cubic polynomials respectively indicating the quadratic provides better performance. The proposed algorithm also establishes the constancy in gait speed estimation which can be observed as a decrease in standard deviation of absolute error from 0.15 m/sec to 0.04 m/sec.