Seat Belt Fastness Detection Based on Image Analysis from Vehicle In-abin Camera
Alexey Kashevnik, Ammar Ali, Igor Lashkov, Nikolay Shilov
Abstract
Seat belt fastness detection in vehicles is an important factor due to the high protection role in case an accident occurs. Modern vehicles usually have belt fastness detection systems that can be simply tricked. There are also algorithms that can recognize seat belt fastness based on driver visual monitoring. Unfortunately, the existing algorithms are not so efficient and car manufactures do not implement them to vehicles. Most of them based on Hough, Canny, or other edge detection. In this paper, classification for driver seat belt status using a camera inside the driver cabin is proposed. The model based on YOLO neural network for detecting the driver seat belt fastness. Two steps approach was used to solve the problem: the main part of the belt detection and corner detection. These steps allow the system to recognize the situation when the seat belt is fastened behind the human body. Tiny-YOLO was used to detect the main part of the belt as the first object as well as the belt corner as a second object. The model classifies belt fastness between three cases: the belt is not fastened, the belt is fastened correctly, and the belt is fastened behind the back.