With the recent advent of ubiquitous computing and sensor technologies, human mobility data can be acquired for monitoring and analysis purposes, e.g., Daily routine identification. Mining mobility data is challenging due to the spatial and temporal variations of the human mobility, even for the same activity. In this paper, we propose a methodology to first summarize indoor human mobility traces as a flow graph using a probabilistic grammar induction algorithm. Then, we recover salient mobility patterns as sub flows in the flow graph. Thus, such patterns/sub flows are expected to be corresponding to the activities that often last for a while, e.g., Cooking and cleaning. The weighted kernel k-means algorithm is adopted for the sub flow extraction. Finally, we detect the occurrences of the sub flows along the mobility traces and obtain their daily routines via the eigen-decomposition. To evaluate the effectiveness of the proposed methodology, we applied it to a publicly available smart home data set containing digital traces of an elder living in a smart house for 219 days. We illustrate how the flow graphs, sub flows and daily routine patterns can be inferred from the mobility data. Our preliminary experimental results show that the proposed approach can detect sub flows which are more specific in terms of their correspondence to activities when compared with a frequent pattern clustering approach.