<p>Hazy conditions in videos often degrade visual quality, impacting object detection and scene understanding in real-world applications. Existing video dehazing approaches suffer from inefficiency, poor generalization, and the inability to preserve fine details under varying haze densities. To address these challenges, this study proposes an Attention-Based Global-Local CycleGAN with Enhanced Dragonfly Optimization (EDO) for efficient video dehazing and haze detection. The proposed model integrates a CycleGAN framework that combines global and local attention mechanisms to capture spatial and contextual information, ensuring precise haze removal. The Enhanced Dragonfly Optimization (EDO) is employed to optimize critical hyperparameters, enhancing the convergence speed and accuracy of the model. Wavelet-DNN is used to preserve high-frequency information, ensuring sharper and more detailed reconstructed frames. Experimental evaluations on benchmark datasets demonstrate that the proposed method outperforms existing approaches in terms of PSNR, SSIM, and processing time. The results validate the effectiveness of the model in restoring high-quality, haze-free videos while maintaining real-time performance. The proposed framework offers a robust and efficient solution for video dehazing and haze detection, improving the reliability of video analytics in challenging environments.</p>