As urban tunnels shift from "subjective experience design" to "data-driven design", the visual perception of drivers has become a critical criterion for evaluating safety and comfort. To reveal the differences in urban tunnel landscapes under static and dynamic perception conditions, and to enable quantitative identification and classification of key landscape elements, the authors construct a landscape element extraction method that integrates image recognition and subjective perception analysis. Because conventional urban tunnel landscape designs are mostly based on static perspectives and do not accurately reflect drivers′ real visual experiences during motion, the authors collect dynamic perception data by organizing real driving tests. Image processing technology is used to extract features and identify "blobs" from the data images, thereby distinguishing static/dynamic visual features of the tunnel landscape. The identified landscape elements are then classified and rated using the Schroeder primary-secondary classification method to separate the main landscape elements and the auxiliary ones. The results show that under dynamic driving conditions, drivers focus more on key landscape elements on both sides of the driving path and within overlapping visual fields. Appropriate use of the color and patterns in these areas significantly enhances visual comfort and safety. Under static driving conditions, drivers observe more details; however, these details are difficult to perceive in actual driving and therefore do not contribute to regulating visual recognition perception.