UAV-Based Vehicle Detection by Multi-source Images

Detecting vehicles from autonomous unmanned aerial vehicle (UAV) systems is attracting the attention of more and more researchers. This technique has also been widely applied in traffic monitoring and management. Differing from the other object detection frameworks which just use data from a single source (usually visible images), we adopt multi-source data (visible and thermal infrared images) for a robust detection performance. Since deep learning techniques have shown great performance in object detection, we utilize "You only look once" (YOLO), which is a state-of-the-art real-time object detection framework for automatic vehicle detection. The main contributions of this paper are as follows. (1) Through integrating a thermal infrared imaging sensor and a visible-light imaging sensor on the UAV, we build a multi-source data acquisition system. (2) The rich information from the multi-source data is fully exploited in the proposed detection framework to further improve the accuracy of the detection result.