Drone for Intelligent Traffic Monitoring: Current Status and Future Trends

With a large number of observation objects and diverse scene environments, limited fixed monitoring points and views make the traffic monitoring a complex and difficult task. UAVs, as mobile and agile vehicles, provide an aid for the dynamic implementation of traffic monitoring. Similar to many other application areas, the development of intelligence is mainly driven by deep learning. This paper reviews how UAVs can use deep learning methods for dynamic traffic monitoring. In particular, for detection and recognition, which fall under the umbrella of computer vision techniques, several high-performance methods are brought together, typically represented by the Yolo family of algorithms. For navigation and localization, deep learning methods can be combined into planning and scheduling as an overall control scheme for monitoring a swarm of drones. As for fusion and analysis, the detection results will be abstracted into real-time traffic flow states. It will also be integrated with the weather, time, and other related information, and heterogeneous data processing will be performed using deep learning methods to achieve intelligent analysis and decision-making. In addition, the UAV simulation platform is introduced to address the problem of insufficient actual training data. The types, performance, and typical application cases of traffic monitoring drones are also summarized and explained. With the continuous development of technology, the three aspects involved in dynamic traffic monitoring will form a more functional and cohesive closed-loop system with UAVs as the driving platform and intelligent processing algorithms as the core technology.