Automated Stride Detection from OpenPose Keypoints Using Handheld Smartphone Video

Gait analysis is important for assessing neurological disorders, but existing methods require human assistance and specialized tools, which can be time-consuming and resource intensive. Healthcare providers would benefit from automated gait analysis. The first step in this process is detecting strides and identifying foot events. While many studies have focused on identifying strides, few have utilized handheld videos. In this study, walking gait was recorded using a handheld smartphone at 60Hz. Body keypoints were identified using the OpenPose - Body 25 pose estimation model, and a new algorithm was developed to identify the movement plane, foot events, and strides from the keypoints. The stride identification results were compared to ground truth foot events labeled through direct observation. Stride detection was accurate within 2 to 5 frames, demonstrating the viability of automated stride detection for smartphone-based motion analysis. This new approach can help improve gait analysis to be more efficient and accessible, especially for individuals who require remote monitoring.