Markerless shank and foot segmentation and skeletonization for lower limb 2-D movement analysis during the stance phase in gait

A method for analyzing the 2-D movement of the shank and foot complex during stance is presented. A single camera acquiring at 50 frames per second was positioned laterally to the subject in order to obtain an approximate sagittal view of the shank and foot during the stance phase. Experiments were performed in two conditions: with the subject barefoot and wearing socks. Shank and foot were first separated from the background and then shank, rear-foot and fore-foot were identified. The background/foreground separation (segmentation) was achieved by using Mixture of Gaussians (MoG) method while the lower limb segment definition (skeletonization) was done by using Medial Axis Transform (MAT). The results of the segmentation showed that the presence of socks did not affect the performance of the segmentation algorithm. The skeletonization showed that a three-segmented model could successfully be fit to the portion of the lower limb under analysis. For each frame of the stance phase, three lines representing the three segments could be fit to the skeletonized images and therefore a full 2-D movement analysis could be obtained. The proposed method not only represents a good alternative to marker-based methods but also offers at least two potential advantages: the option of wearing socks during the trials and the possibility of describing the relative movement between fore-foot and rear-foot.