Online cross camera vehicle tracking in high-speed tunnel highway scenarios is particularly challenging because drastic illumination changes cause severe appearance instability, while practical deployment also requires low computational cost. To address these challenges, we propose a cross camera online tracking framework based on constrained cost association and an online steady state mechanism. The global association is decomposed into a series of constrained subproblems over adjacent camera pairs and lane groups derived from road topology and lane semantics. For each subproblem, we construct a multi-constraint cost matrix and use the Hungarian algorithm to obtain consistent one-to-one assignments, avoiding heuristic greedy matching. To further address severe illumination changes and unstable appearance features, the online steady state mechanism maintains a robust identity representation through quality-driven updates with outlier suppression. Experiments on a real tunnel highway dataset demonstrate that the proposed method substantially outperforms representative baselines in terms of IDF1, MOTA, ID switch count and FPS , achieving 14.93 FPS for practical intelligent transportation applications.