Abstract: We present a novel approach that combines the snake model with a 3D segmentation technique to segment potential pedestrians from traffic images and extract their contour curves. The snake model is highly prone to interference by noise, and thus, an anti-noise snake model is proposed that combines the gradient vector flow (GVF) model with a tailored corner detection approach. In addition, an optimization method for the initial contour curve and a modeling method for the adaptive GVF field are proposed to overcome problems associated with establishing the initial contour curve, given the high computational complexity of the GVF model during an iteration process and other considerations. This optimization method reduces the time required to effectively extract contour curves. Finally, using pedestrians in complex traffic scenarios as objects, the performances of the GVF model and anti-noise snake model are tested and compared for extracting pedestrian contours. Results show that the anti-noise snake model is effective for extracting pedestrian contour curves in complex traffic scenarios.