Identificação da utilização de cinto de segurança em condutores de automóveis através de visão computacional

Abstract

The constant growth in the number of vehicles on the streets has led to an increased risk of traffic accidents and consequently, the number of deaths caused by these accidents. Historically the use of seat belts has proven to be effective in preventing such deaths, but some people do not use them due to lack of habit or negligence. The existing automated mechanisms for traffic monitoring provide only limited information, which do not include the use of the seat belt. Due to the limitations of these mechanisms, there is a demand for strategies of traffic surveillance, control and management that are more intelligent and efficient, and the field of Computer Vision has been considered a viable alternative in the development of solutions that can meet this demand. This dissertation describes the research procedure used in the development of a software prototype to aid identification of the wearing of the seat belt among car drivers. Initially, real traffic scenes were captured on video, and features of interest were extracted using Digital Image Processing and Computer Vision techniques. To evaluate the accuracy of the system, the results were compared with those obtained by human evaluation, achieving good results under certain environmental conditions