Abstract: In the tower area, traditional unmanned aerial vehicle inspection methods that rely on absolute coordinates and preset routes are prone to significant positioning drift, delayed obstacle response, and increased blind spots in shooting, resulting in insufficient obstacle avoidance capabilities and limited coverage. Therefore, this study proposes a rapid planning method for real time kinematic (RTK) unmanned aerial vehicle (UAV) tower inspection routes based on visual task guidance. Real time collection and processing of tower images through drone vision system, combined with feature extraction and matching technology to accurately identify inspection targets and suppress misidentification and missed detections. Integrating RTK module to achieve high-precision dynamic positioning and compensate for coordinate deviation under electromagnetic interference; Under the guidance of visual tasks, the shooting points and routes are dynamically planned based on real-time image analysis results, with visual recognition areas as spatial constraints. By connecting the shooting points and adjusting real-time collisions, a comprehensive and effective obstacle avoidance route is generated. In the experiment, the flight control accuracy is in horizontal direction ±(10 mm+1 ppm) and in elevation direction ±(15 mm+1 ppm). It's found that the minimum deviation of the route planning using this method is only 0.5 m, the safety index remains above 0.8, and the inspection coverage rate reaches 99.8%. This indicates that this method significantly improves the planning robustness and inspection efficiency in harsh tower environments. Compared to traditional methods with a maximum deviation of 9 m, this method significantly suppresses positioning drift under electromagnetic interference, ensuring the safety, reliability, and coverage of near tower inspections, and meeting the practical requirements of centimeter level accuracy for power inspections.