Muscle functions are closely related to the architectural characteristics which can be investigated in vivo by ultrasonography. However, it is a challenging problem to automatically extract the architectural parameters, such as cross-sectional area (CSA), from the ultrasound images. In this study, a novel algorithm was proposed to automatically detect the muscle boundary and compute the continuous CSA change from the image sequence. This algorithm was then applied to study the in vivo behavior of the rectus femoris (RF) muscle during isometric contraction. The results showed that the algorithm was highly reliable, and it can provide the information about the muscle contraction from the morphological aspect with high temporal resolution.