DATASET DESIGN FOR DEVELOPMENT OF YOLOV4-BASED OBJECT RECOGNITION MODEL FROM TRAFFIC SIGN RECOGNITION SYSTEM FOR AUTONOMOUS TRAM

• Main Author: Silmy Taliyan, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/67896
• Institution: Institut Teknologi Bandung

One alternative to electric-powered public transportation that can be use by residents as a mode of transportation in the city is the tram. Trams in operation move in mixed traffic conditions so that the possibility of accidents with other road users also higher. The insensitivity and negligence of road users (including trams) to traffic conditions symbolized by traffic signs is generally one of the causes of accidents. Efforts to overcome this accident are carried out by developing trams that can operate autonomously based on the operating traffic conditions which are depicted/symbolized by traffic signs. This capability is realized in a system called the traffic sign recognition system. This system is built with an artificial intelligence model to recognize traffic signs. In making the model, it takes data that can be a reference to be learned by the model so that it can operate as desired. The data is collected and annotated so that a dataset is formed that will be used for model training as a process of making object recognition models. Dataset development also requires certain criteria so that the resulting model can represent the desired performance. Variation of data is needed so that the resulting model does not overfit. This is done by ensuring that the referenced data are not similar or exactly the same as each other, augmentation can also be done for conditions that were not found in the initial data. With the dataset created, the model is expected to operate as desired with good performance.