Gait Event Recognition Based on Edge Detection and Optical Flow from Foot Video

In this study, we proposed a method for extracting gait features from foot video that takes into account the privacy of pedestrians. Here, frames in which heel-strike and toe-off occurred were detected from the gait video for the purpose of gait feature extraction. First, edge detection and optical flow calculation were performed on a video of a pedestrian's feet. Subsequently, these were used to extract areas of motion, and clustering was performed using the distance information in the image to extract candidate pedestrian feet. Next, acceleration flow was computed, and frames in which heel-strike occurred were detected by analyzing the flow within the candidate of the pedestrian's feet. In addition, edge detection and optical flow were used to search for the static regions, which were then used to detect frames in which toe-offs occurred. The toe-off detection method in this study is simple. However, it is promising because it utilizes human gait characteristics. Finally, several experiments were conducted to evaluate the method proposed in this study. Experimental results showed that both heel-strike and toe-off were detected with high accuracy. False positives and undetected cases were observed, but all were potentially addressable. Furthermore, there was no significant difference in the results of both heel-strike and toe-off detection between a cluttered background and a tidy background. These results confirm the effectiveness of this method.