Seat Belt Detection Using Genetic Algorithm-Based Template Matching

Abstract

In this study, we focused on detecting seat belts of drivers from camera images of the front windshield area of cars. When the color of the seat belt differs from that of the clothing worn by the driver, the seat belt can be detected by focusing on edge and pixel value differences. This study created a template representing pixel value differences and applied template matching using a genetic algorithm to achieve higher detection rates than edge-based methods. To quantitatively evaluate the performance, we created a dataset consisting of 73 images and compared the performances of each method. Notably, the proposed method achieved higher detection rates than the other methods.