Adaptive Video Text Tracking Based on Pixel-level Feature Extraction

The increasing utilization of video text has made text detection and tracking in videos an important research direction in the field of computer vision.To address the issue of existing methods being inaccurate at detecting curved text, an adaptive video text tracking (ATDTF) model based on pixel-level feature extraction was proposed.First, pixel-level information was used for text detection, and then multi-directional prediction and connected component analysis were performed to achieve the accurate detection of text in complex scenes.Next, the tracking algorithm was used to associate and track text targets in consecutive frames.Finally, the target motion model or appearance model, in combination with Kalman filtering, was used to associate and predict the trajectory to achieve accurate tracking of the target.Experimental results show that, the ATDTF model improves the F1 value by 0.6% over the current method, the detection speed is at least 21 frames per second faster, and the running speed is increased twofold.The proposed model achieves performance improvement in video text detection and tracking tasks, providing an effective end-to-end solution for video text information processing.