Intersection Performance Evaluation Based on Indonesian Road Capacity Manual with Video Analytics

Intersections play an important role in traffic flow management, and their performance analysis is different from that of roads. Indonesian Road Capacity Manual 1997 or in Indonesia called MKJI provides guidance for evaluating intersection performance by assessing vehicle movement between intersection arms using the Turn Movement Count (TMC). This study investigates the use of video analytics to assess intersection performance using the MKJI 1997 framework. The system is organized into three layers: perception, analytics, and business. In the analytics layer, video analytics is implemented using the YOLOv8s model, which has the greatest Recall and F1-Score of 0.85 and 0.919 when compared to other models. This is combined with the Bytetrack tracker, which outperformed Deepsort in MOTA tests across three scenarios. The video analytics system computes the TMC for each intersection leg, and the Sedayu intersection in Yogyakarta was tested over two days (weekdays and weekends) at three different time intervals. The findings revealed that the intersection has a mediocre Service Level, scored E or F in all time intervals. Several enhancements can be made to increase capacity based on the MKJI 1997 standards, and future research could concentrate on minimizing signal cycle times and developing more efficient vehicle recognition models with real-time, high-FPS capabilities for precise tracking.