Deep Neural Network-based Detection of Road Traffic Objects from Drone-Captured Imagery Focusing on Road Regions

This paper presents a novel deep learning approach for the detection of traffic objects from drone-based imagery, focusing predominantly on the rapid and accurate detection of vehicles within road sections. The proposed method consists of two primary components: a road segmentation module and a vehicle detection network. The former leverages a residual unit with skip-connections to effectively extract road areas, while the latter employs a modified version of the YOLOv3 architecture, tailored for high-accuracy and high-speed vehicle detection. To address the issue of data imbalance, which is a pervasive challenge in drone images, this paper utilizes a range of data augmentation techniques to improve the robustness of the proposed model. Experimental results on the UAVDT and UAVid datasets exhibit that the proposed model attains a substantial boost in accuracy and inference speed of vehicle detection in comparison to the existing methods. These findings underscore the potential of the proposed approach for real-world traffic monitoring applications, where rapid and reliable vehicle detection is paramount.