Accuracy of Kinovea software in estimating body segment movements during falls captured on standard video: Effects of fall direction, camera perspective and video calibration technique

Abstract

Falls are a major cause of unintentional injuries. Understanding the movements of the body during falls is important to the design of fall prevention and management strategies, including exercise programs, mobility aids, fall detectors, protective gear, and safer environments. Video footage of real-life falls is increasingly available, and may be used with digitization software to extract kinematic features of falls. We examined the validity of this approach by conducting laboratory falling experiments, and comparing linear and angular positions and velocities measured from 3D motion capture to estimates from Kinovea 2D digitization software based on standard surveillance video (30 Hz, 640x480 pixels). We also examined how Kinovea accuracy depended on fall direction, camera angle, filtering cut-off frequency, and calibration technique. For a camera oriented perpendicular to the plane of the fall (90 degrees), Kinovea position data filtered at 10 Hz, and video calibration using a 2D grid, mean root mean square errors were 0.050 m or 9% of the signal amplitude and 0.22 m/s (7%) for vertical position and velocity, and 0.035 m (6%) and 0.16 m/s (7%) for horizontal position and velocity. Errors in angular measures averaged over 2-fold higher in sideways than forward or backward falls, due to out-of-plane movement of the knees and elbows. Errors in horizontal velocity were 2.5-fold higher for a 30 than 90 degree camera angle, and 1.6-fold higher for calibration using participants’ height (1D) instead of a 2D grid. When compared to 10 Hz, filtering at 3 Hz caused velocity errors to increase 1.4-fold. Our results demonstrate that Kinovea can be applied to 30 Hz video to measure linear positions and velocities to within 9% accuracy. Lower accuracy was observed for angular kinematics of the upper and lower limb in sideways falls, and for horizontal measures from 30 degree cameras or 1D height-based calibration.

Do It by Yourself: Larval Locomotion in the Black Soldier Fly Hermetia illucens, with a Novel “Self-Harvesting” Method to Separate Prepupae

“…This difference may lay in the modification of the mouthparts during the transition between the 6th instar and the prepupae: the first have functional mouthparts, which are severely reduced in the latter [26,27]. To the best of our knowledge, ours is the first use of the software Kinovea to describe the movement patterns of an insect, while it is employed in other different research contexts [59][60][61][62], encouraging further applications to study the movements of other animal species [63][64][65][66][67].…”

Injuries from falls by older adults in long-term care captured on video: Prevalence of impacts and injuries to body parts

“…Finally, we examined how injury risk associated with the occurrence of impact to body parts, but we did not measure the severity of impacts, based for example on impact velocity. Further work is needed to understand how injury risk depends on detailed kinematic characteristics of falls [ 61 , 62 ].…”

Identification of Glucose Transport Modulators In Vitro and Method for Their Deep Learning Neural Network Behavioral Evaluation in Glucose Transporter 1–Deficient Mice

“…However, we expanded the tracking to multiple body parts to analyze more subtle changes that the drugs may potentially cause. Because of the ataxic locomotion of G1D mice, multi-point tracking of mouse body parts proved imprecise using Kinovea (Shishov et al, 2021). Thus, we used a deep neural networks for completely automated pose estimation.…”