Uncalibrated Single-Camera View Video Tracking of Head Impact Speeds Using Model-Based Image Matching

This study evaluates the accuracy of a model-based image matching (MBIM) approach with model calibration for tracking head impact speeds in uncalibrated spaces from single-camera views.

Two validation datasets were used. The first included 36 videos of guided NOCSAE headform drops at varying camera positions (heights, distances, camera angles) where a speed gate measured vertical impact speed. The second dataset had eight videos of participants performing ladder falls with marked helmets, captured using a 12-camera motion capture system to track head impact speeds. Each video was tracked frame-by-frame, matching a 3D NOCSAE headform model to the head using MBIM software. Accuracy was assessed by comparing captured to MBIM-tracked speeds by the mean difference and Root Mean Square Error (RMSE). A linear model assessed the influence of camera position.

For ideal camera views (90 degrees, height 1 or 1.4 m), MBIM-tracked vertical speeds were 0.04 ± 0.15 m/s faster than the true speed (RMSE 0.15 m/s; 2.3 ± 6.2% error). Across all 36 NOCSAE videos, MBIM-tracked vertical speeds were 0.03 ± 0.19 m/s faster (RMSE 0.19 m/s; 1.8 ± 6.9 % error). In participant videos, MBIM-tracked resultant speeds were 0.01 ± 0.33 m/s slower (RMES 0.31; 0.7 ± 9.5% error) compared to motion capture.

MBIM with model calibration can analyze head impact kinematics from single-camera footage without environment calibration, achieving reasonable accuracy compared to other systems. Analyzing head impact kinematics from uncalibrated single-camera footage presents significant opportunities for assessing previously untraceable videos.