Aiming at contributing to the development of a robust computer vision traffic surveillance system, in this work a method for vehicle identification and tracking that applies the Scale Invariant Feature Transform (SIFT) and a Kalman filter is proposed. The SIFT algorithm extracts keypoints of the moving object on a sequence of images and the Kalman Filter provides a priori estimates of vehicle position and velocity which are used to improve the said algorithm. This strategy allows reducing the amount of pixels to be tested for matches within the whole image scenario by dynamically redefining the ROI (Region of Interest). Using algorithms from OpenCV Library to compose the required computer vision tracking method, a prototype system was constructed and submitted to off-line experiments based on a series of grabbed traffic image sequences. From the results, it is possible to assert that the joint use of SIFT and Kalman filtering techniques is able to improve the overall algorithm performance concerning quality of matches between the images of the object and the scene, since it reduces in 50% the total number of false positives, one the main limitations of the pure SIFT algorithm.