Abstract: Intelligent transportation systems (ITSs) have been widely used in smart cities with a widespread problem of missing sensing data. The limited storage computing capability of traffic stations also severely restricts the recovery of sensing data and greatly affects the normal use of ITSs. Although the powerful computing capacity of edge nodes can be used to alleviate this issue, the high complexity and dynamics of the temporal and spatial correlation of sensing data still pose a serious challenge to the recovery process, making the result of data recovery, based on static edge nodes deployment and distribution, unsatisfactory. In order to effectively solve this series of problems, we propose an adaptive urban traffic sensing data recovery system based on intelligent edge computing. The system mainly consists of two parts: Firstly, the submodular optimization theory is used to design a suboptimal deployment and allocation scheme for edge nodes with a theoretical performance lower bound. Secondly, we address a data recovery method based on the low-rank theory. At the same time, the recovery results are used to calculate the corresponding non-missing theoretical lower bound, feed back to the edge nodes, and then update the data distribution scheme to ensure an accurate recovery of subsequent sensing data. The experiments based on large-scale ITSs traces of Australia illustrate that our method can achieve 90% of the optimal performance for the edge node deployment, and improve the data recovery accuracy by 43.8% in comparison with three baselines. Furthermore, the adaptive data recovery scheme can improve the accuracy by 40.3% on average.