VEHICLE TRAFFIC VOLUME COUNTING IN CCTV VIDEO WITH YOLO ALGORITHM AND ROAD HSV COLOR SPACE BASED SEMANTIC SEGMENTATION SYSTEM DEVELOPMENT

• Main Author: Stanley, Abel
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/56070
• Institution: Institut Teknologi Bandung

Traffic congestion is a significant problem in developing countries. One viable solution is a Smart Traffic Light System which utilizes artificial intelligence to adapt light configuration to actual traffic condition in real time. To adapt properly, the system would need traffic density information. We propose a vehicle counting system with neural networks to calculate vehicle volume in traffic roads. In the proposed system, a vehicle is detected with YOLO (You Only Look Once), the state-of-the-art of neural network-based object detection algorithm. The model’s performance can be improved with the extraction of RoI (Region of Interest), which is traffic roads. RoI extraction is implemented with HSV color model based semantic segmentation. Vehicle detection is followed by vehicle tracking and counting. Three tracking algorithms are experimented with the system: KCF (Kernelized Correlation Filter), CSRT (Channel and Spatial Reliability Tracking), and MOSSE (Minimum Output Sum of Squared Error). Vehicle counting is implemented in two methods: incremental or actual. A graphical user interface is developed to provide easy access to system configurations. The result reveals that the best system configuration in terms of accuracy while capable of running in real-time for CCTV recordings is YOLOv4 (608x608) with KCF tracker and RoI Extraction.