Robust lane marking detection remains a challenge, particularly in temperate climates where markings degrade rapidly due to winter conditions and snow removal equipment. In previous work on stereo images, dynamic Bayesian networks with heuristic features were used whose distributions are identified using unsupervised expectation maximization, which greatly reduced sensitivity to initialization. This work has been extended in three important respects. The situations where poor RANSAC hypotheses were generated and significantly contributed to false alarms have been corrected. The null hypothesis is reformulated to guarantee that detected hypothesis satisfy a minimum likelihood. The computational requirements have been reduced for tracking and pairing by computing an upper bound on the marginal likelihood of all part hypotheses and rejecting part hypothesis if its upper bound is less likely than the null hypothesis. Therefore, parts that could never surpass the null hypotheses are excluded from contributing to the n^2 complexity of the tracking and pairing processes. These improvements have led to real-time operation at the frame rate of the stereo camera, and robust performance improvements on our dataset. The dataset collected in the region of Michigan which has significantly degraded lane markings have been evaluated for this work.