To realize the assistance of transrectal ultrasound imaging in robotic prostatectomy, registration between ultrasound equipment and surgical robotic system needs to be established. The recognition and segmentation of minimally invasive surgery (MIS) instrument contour in ultrasound image is an important part of registration process, and its accuracy will directly influence the registration results. An improved parametric active contour algorithm (Snake) under global constraints is proposed, which can accurately segment surgical instrument contour in ultrasound images. The improved Snake method uses known instrument profile as prior knowledge, describes the information in normalized Fourier descriptors, and applies it in Snake external energy to constrain the segmentation result gradually near the desired contour. During the solving process, the Jacobian matrix of energy function relative to parameters is calculated to acquire the partial differential equation with minimum energy. Then the partial differential equation is discretized, deduced with Euler method to obtain segmentation result. The experiment shows that when using Snake method with prior knowledge, the segmentation result of surgical instrument contour in ultrasound image can be realized with high accuracy, and the contour result is similar with real instrument profile. Compared with traditional Snake method, the improved method constrains curves globally, and can avoid contour curve falling into local minimum, having a great application prospect.