Development of a Vehicle Classification for Multi Lane Free Flow Using the EfficientDet Model

Transportation plays an important role in daily life, supporting economic, social, and logistical activities. The increase in the number of vehicles in Indonesia, which reached $148,212,865$ units in 2022 (BPS), has led to significant congestion, especially at toll gates. One of the contributing factors is that the vehicle classification process is still conducted manually using infrared sensors combined with officer observation. This method is not efficient, prone to human error, and heavily dependent on human resources. To address this problem, this research develops an automated vehicle classification system using a Deep Learning Convolutional Neural Network with a pre-trained EfficientDet model. The system aims to improve accuracy, reduce processing time, and enable real-time operation. In testing with a minimum threshold value of 0.4, the model achieved a True Positive (TP) value of 141 out of 175 data (80.57%). It also demonstrated real-time inference capability, processing at 8 frames per second (FPS).