Accurate characterization of rock mass discontinuities is critical for stability analysis and support design in underground tunnel construction. While three-dimensional point cloud technology offers high-precision geometrical data, traditional identification methods predominantly focus on isolated tunnel faces, neglecting the longitudinal spatial correlations inherent in the continuous excavation process. This study introduces a dynamic identification framework that leverages sequential tunnel face point clouds to enhance discontinuity recognition. By utilizing previously excavated faces as prior spatial information, we employ a sequential Kalman filtering method with adaptive measurement noise to predict, associate, and refine the discontinuity parameters of the current face. The discontinuity orientations are represented using unit normal vectors to avoid angular discontinuities at the 0°/360° boundary, while a cosine-distance-based matching strategy is used to associate discontinuity sets between consecutive tunnel faces. This approach effectively distinguishes persistent geological discontinuities from random, blast-induced fractures, significantly reducing false positives and improving the stability of discontinuity orientation estimates. The proposed framework is validated using 26 consecutive tunnel faces from a tunnel in Tibet, China. The results demonstrate that the sequential updating process effectively suppresses fluctuations in the identified discontinuity parameters and captures the spatial evolution of persistent discontinuity sets along the tunnel alignment. The proposed framework facilitates the construction of Discrete Fracture Network (DFN) models and supports accurate rock mass classification. This research contributes to the development of geotechnical digital twins by providing a continuous, evolving geometric representation of the tunnel environment, ensuring safer and more efficient construction management.