Sistem Deteksi Lampu Lalu Lintas Sebagai Asisten Pengemudi Menggunakan Convolutional Neural Network

Abstrack - Accident cases in Indonesia are increasing along with the increase in the number of motorized vehicles. From 2016 to 2017, speed limit violations increased by 96.20% and violations of road markings or signs also increased by 5.54%. Intelligent transportation system is one solution to reduce the number of accidents. Currently Driver Assistance Systems (DAS) are being developed in the automotive world. The purpose of this research is to design a watershed based on three input parameters for determining recommended actions, namely: 1) distance to the vehicle behind; 2) vehicle speed; and 3) traffic light status with recommendation action using fuzzy rule base. Lidar sensor for distance detection and GPS for monitoring vehicle speed. The YOLOv4 Algorithm method is used to detect traffic lights. The results of this study, the accuracy of sign color recognition is 92.831% with a detection speed of up to 8.94 FPS. The most stable reading distance is between 1 – 8-meters with a light intensity of 10 – 3200 lux and a tilt angle of up to 90 degrees. There is a drop in processing speed of up to 1.5 FPS during system integration. This DAS is effective enough to be applied to two-wheeled and four-wheeled motorized vehicles.