The TRACLUS (trajectory clustering: a partition and group framework) algorithm's reliance on purely spatial metrics for its partitioning and clustering phases leads to inaccurate segmentation and a failure to identify coupled spatio-temporal patterns, especially with noisy, density-varied data. To address these limitations, we propose a trajectory clustering algorithm based on spatio-temporal weighting and density peak (STW-DP-TRACLUS). This method enhances the TRACLUS in three key aspects: First, spatio-temporal weighted segmentation integrates adaptive weights and geometric distances into the minimum description length cost function to extract sub-trajectory segments with local features; Second, improved density peak clustering incorporates spatio-temporal local density, novel noise identification, and iterative center selection to boost clustering quality; Third, density filtering and continuity constraints are embedded in the representative trajectory generation to enhance interpretability. Experiments on taxi datasets from Beijing and Shanghai, featuring different network topologies, demonstrate that the proposed method alone reduces reconstruction error by 28.15% and direction deviation by 66% on average. Overall, the STW-DP-TRACLUS outperforms traditional and state-of-the-art algorithms across three evaluation metrics, validating its efficacy in mining complex spatio-temporal patterns.