<p>Detection and recognition of text in video content have become crucial research areas in computer vision due to the rapid expansion of multimedia data and the growing demand for automated information extraction. The dynamic nature of video introduces challenges such as varying lighting conditions, motion blur, complex backgrounds, and text appearing at different scales. This paper explores recent deep learning techniques for detecting and recognizing Arabic text in videos, focusing on enhancing the precision of text localization. We review state-of-the-art methods, particularly YOLO (You Only Look Once) and RCNN (Region-based Convolutional Neural Networks), for text detection. In addition, we investigate fine-tuning techniques and the U-Net deep model to improve the accuracy and robustness of text detection and recognition in video frames, utilizing CNN (Convolutional Neural Network) architectures for character recognition. Our research includes a comparative analysis of these methods with alternative approaches, incorporating temporal information from consecutive frames to enhance text consistency and recognition accuracy using Gabor filters. Experimental results show the superiority of YOLO in text detection and the U-Net model in character recognition.</p>