<![CDATA[Concentration are crucial when driving. Drivers who lose their concentration tend to have a slower reaction time, and a higher possibility of violating traffic signs. Traffic signs violation is considered a criminal act with harsh penalties. In addition, traffic sign violations interferes with comfort and endanger other road users. Therefore, we need a system that is able to detect signs accurately and quickly which can inform driver in advance. A research on traffic signs detection on Swedish and Slovenian traffic signs use Mask R-CNN model which based on convolutional neural networks [18]. These method was capable of achieving a mAP@50 score that exceeds 95%. However, the research did not evaluate on the detection speed of such methods. In this research, YOLOv4-tiny is used to detect Indonesian traffic signs. Dataset used in this research are independently collected, which consist of nine prohibition signs and two command signs. The YOLOv4-tiny method with input size of 416 x 416 is able to achieve mAP@50 score of 88.55% with detection speed of 19.41 FPS. With modification to input size and dataset, YOLOv4-tiny are able to achieve mAP@50 score up to 89.58% and detection speed up to 30.87 FPS. YOLOv4-tiny are also able to detect road signs from distance of around 5 to 15 meters with 80.42 % accuracy. Indonesian traffic sign recognition program made by utilizing the YOLOv4-tiny model achieve average recall of 72.9%.]]>
