This paper describes the procedures adopted at the research and implementation of a software prototype used to assist the identification of the seat belt in car drivers. We applied computer vision techniques for the preparation and extraction of elements of interest from the videos and pictures. In some specific cases, the use of the seat belt was imperceptible even when evaluated by the human eye and in such situations it was proposed to use a Bayesian network that employs characteristics of the image to assist the identification. Experiments performed under controlled environmental conditions showed that the prototype achieved promising results.