Traffic Congestion Detection Using Fixed-Wing Unmanned Aerial Vehicle (UAV) Video Streaming Based on Deep Learning
Winahyu Utomo, Putu Wisnu Bhaskara, Arief Kurniawan, Susi Juniastuti, Eko Mulyanto Yuniarno
Abstract
Population growth in the region has led to increased use of roads that causes traffic congestion. Traffic congestion also occurs during long weekend in outsides city. The roads in the area are usually smooth traffic flow, it becomes very congested. One method of the smart road monitoring system is a fixed camera sensor as input and artificial intelligent as the analysis. However, these methods require a great infrastructure on highways such as: power supply, protective CPU, good power supply and stable computer network connections. This will be difficult to fulfill if applied on roads outside the city. To overcome the problem, we propose a system of vehicles detection and road density classification using Fixed-Wing Unmanned Aerial Vehicle (UAV) video streaming. We chose Fixed-Wing UAV for its advantages: wide range and fast flight speed. The proposed system detects and classifies vehicles. Vehicles are detected and calcified using CNN's Deep Learning which uses the YOLO architecture. The level of traffic density is determined by the area of the road covered by vehicles to road area. We tested the proposed system using YouTube video and UAV video streaming. Both experimental scenarios have almost the same results. Precisions of recording video UAV and streaming video are: 90.75% and 90%, respectively.