Abstract

This research discusses the implementation of traffic monitoring in calculating the level of traffic density based on the level of service (LOS) value of in homogeneous traffic or heterogeneous traffic. According to previous study, object detection's accuracy is frequently questioned in congested traffic situations. YOLO, on the other hand, can consistently detect objects and is ideally suited for traffic density studies. It is feasible to determine traffic density using a combination of SVR and occupancy area. The total number of training data used by the YOLO method to detect vehicle types was 4665 samples of vehicles consisting of types 1-6. The SVR method uses variables processed using basic freeway segment for training data and occupancy area for test data. The results show that the YOLO method can recognize vehicle types and obtain 75.16% accuracy in daytime traffic conditions. For the estimate of traffic density based on occupancy the YOLO and SVR method is implemented. This is represented with a polynomial kernel with epsilon optimization parameter = 1.0, degree = 1, gamma = 0.0, and coef0 = 2.0 obtaining a MAPE score of 53.59; this value is smaller than the use of a linear kernel getting a MAPE value of 55.5.