The traffic problem has several components that may be discussed: vehicles, pedestrians and the interaction between them. This paper proposes a method for acquisition of pedestrian traffic variables, using computer vision techniques. Isolated pedestrians, groups of pedestrians and vehicles at the scene are detected from a video sequence, using a background model. Pedestrians are tracked on the image using their shape and optical flow. Counting is done on any area of the scene to estimate the flow and direction of movement. The proposed method can be configured under different perspectives from a set of examples. The experimental results on crosswalks show that this method allows estimating the variables of interest in complex scenes.