Pengembangan Aplikasi Deteksi Potensi Pindah Lajur Kendaraan Menggunakan Yolo V4

Abstract

Driver factors need attention because they tend to cause traffic accidents. Driver factors plus the lack of safety features such as LDWS and ADAS inside vehicles in Indonesia are the reasons for this research. This study proposes a driver assistance application that can detect lane departure movement of a vehicle. If done with a camera, the application might help minimize driving negligence. YOLO v4 is selected to meet application needs such as high detection accuracy and fast computation time. The YOLO v4 model was trained using a dataset containing 400 images and obtained mAP value of 42,98% with FPS value range of 18,7 FPS to 40,5 FPS. Based on test results, the application can detect driving lanes well and give warnings as the departure movement reaches 20% or more.