Object detection in haze for UAV using DCP and YOLO

The phenomenon of haze poses a significant challenge in UAV (Unmanned Aerial Vehicle) surveillance, particularly under the computational constraints inherent to UAV systems. This paper proposes a sophisticated yet efficient approach to overcome these limitations by integrating algorithms like dark channel prior (DCP), dynamic gamma correction, bilateral filtering, and YOLOv8 object detection. The DCP algorithm effectively clears haze from images by manipulating pixel intensity values, this process is further optimized using bilateral filtering for computational efficiency. These dehazed images are put thorough dynamic gamma correction which further enhances image clarity by dynamically adjusting gamma values for each image. The YOLOv8 architecture, selected for its low computational footprint and superior accuracy, performs object detection and localization on processed images. Using the CVPR 2024 Haze Detection Challenge dataset, the proposed model demonstrates superior performance, achieving a high confidence score and a detection accuracy of 92%, exceeding the capabilities of existing models. These results underscore the method’s effectiveness in addressing haze-related challenges within the computational limits of UAVs.