Real time seatbelt detection using YOLO deep learning model

Abstract

Monitoring seat belt fastening is an essential safety element in automobiles since it increases protection in the event of an accident. Modern vehicles generally have belt fastness monitoring technology that is easily tricked. A seat belt can also be secured using algorithms that rely on the driver's visual monitoring. Sadly, automakers do not include them in their vehicles because the present algorithms are not very efficient. Most of them employ edge detection strategies such as Hough, Canny, or others. In classifying the driver's seat belt status, this article advises installing a camera within the driver's cabin. To ascertain whether the driver's seatbelt is fastened, use the YOLO neural network-based model. Using the corner detection and the main belt detection, the issue was resolved in two steps. The device can detect whether the seat belt is tight behind the body thanks to these motions. Using tiny-YOLO, the first item was determined to be the major piece of the belt, and the second item to be the corner of the belt. According to the model, which classify belt fastness into these two categories, the belt is neither correctly fastened or it is not fastened. This study presents the comprehensive seatbelt dataset that was assembled from various sources for the detector's training, testing, and validation processes. You Only Look Once (YOLO) will enable accurate seatbelt detection in the process.