For development of transportation models, intelligent transportation systems (ITS) provide new opportunities to collect data at a relatively low cost. However, data from ITS sources – such as cell phones, social media applications, GPS, and Bluetooth devices – may not contain the associated activity-based travel patterns. For instance, cell phone data typically consist of locations and timestamps of calls made and received by travelers. From these data, the activity-based travel patterns, such as the activities of the callers, are not clear. In order to capture these latent patterns, the current state of the art focuses primarily on the use of parametric models, which make strong restrictive assumptions about a priori knowledge of the number of underlying activities. To address this limitation, a nonparametric Bayesian estimation approach is proposed to capture latent travel activities without prespecifying their cardinality. The proposed non-parametric model, a hierarchical segmented infinite hidden Markov model, overcomes the need to perform cross-validation arising from standard expectation maximization training or model selection to find the optimum cardinality of hidden activities. The proposed model can summarize large amounts of ITS data that are without activity labels into a few meaningful activity groupings. Then, semantic meaning is added to these groupings by using other rich information, such as land use and time-of-day characteristics. The proposed model is general enough to be applied to any sequential ITS data.