Trajectory similarity measurement faces scalability challenges, with traditional methods incurring high computational costs and many deep learning models requiring substantial resources. To tackle these issues, we introduce TS-CPC, a novel self-supervised framework that adapts Contrastive Predictive Coding (CPC)—applying it for the first time in this domain—to efficiently learn trajectory representations via future state prediction. TS-CPC incorporates two innovative data augmentation techniques: Sliding-Window-based Linear Interpolation (SW-LI) for preserving fine-grained structural details, and Trajectory Detour Point Offset (TDPO) for enhancing robustness against noise and drift. Additionally, Trajectory Motion Enrichment (TME) enhances feature representation by incorporating key movement dynamics, such as speed and orientation, enabling precise modeling of complex motion patterns. Experiments on the Grab-Posisi and GeoLife datasets demonstrate that TS-CPC consistently outperforms both heuristic and learning-based baselines in trajectory similarity tasks. It achieves strong performance across large-scale databases and remains robust under varying sampling rates and noise, highlighting its practical applicability.