Deep Learning-Based Seatbelt Detection Using Yolov8x and Resnet-18 for Automated Vehicle Safety Monitoring
Abstract
The number of individuals killed or hurt in road traffic and injuries can be significantly reduced if more persons buckle up. Seatbelt detection is one of essential task in transportation system. However human monitoring of person - seatbelt use from roadside or in vehicle camera error prone and labour intensive is required to reduce increase. The proposed framework uses a yolov8x model to recognize automobiles and then detects the individuals inside them. A resnet-18 model which pretrained on ImageNet and subsequently with a modified fully connected layer, is employed to classify each person as either wearing a seatbelt (person -seatbelt) or not (person-no seatbelt). The proposed system implemented on the data set, the image data set (Karan Panja). The preprocessing methodology for each data set is different the first dataset using image enhancing functions like clahe and gamma correction and the experimental results show that the suggested system is highly effective, achieving training accuracy of 99.87% , validation accuracy of 96.37% and testing accuracy of 100% within only five epochs.