Haze-Proof AI: Drone-Powered Deep Learning Hits 99% Accuracy in Traffic

Researchers in Coimbatore, India, say they’ve developed a drone-based deep learning system designed to keep traffic monitoring accurate when haze reduces image quality. The framework, called TDE-RYOLO, was described in Neural Computing and Applications by H. Haritha and T. Senthil Kumar. It reports 99.15% overall accuracy for estimating traffic density from drone imagery captured in hazy conditions. The study targets a key weakness of unmanned aerial vehicle monitoring: haze causes washed-out colors, lower contrast and blurred edges, leading vehicle detections to fail. TDE-RYOLO addresses this by preprocessing images with Dark Channel Prior and CLAHE, then using a modified YOLOv9 detector dubbed REP-YOLOv9 to improve robustness under low contrast and occlusion.