Spatiotemporal traffic time series (e.g., traffic volume/speed) collected from sensing systems are often incomplete with considerable corruption and large amounts of missing values, preventing users from harnessing the full power of the data. Missing data imputation has been a long-standing research topic and critical application for real-world intelligent transportation systems. A widely applied imputation method is low-rank matrix/tensor completion; however, the low-rank assumption only preserves the global structure while ignores the strong local consistency in spatiotemporal data. In this paper, we propose a low-rank autoregressive tensor completion (LATC) framework by introducing temporal variation as a new regularization term into the completion of a third-order (sensor $\times $ time of day $\times $ day) tensor. The third-order tensor structure allows us to better capture the global consistency of traffic data, such as the inherent seasonality and day-to-day similarity. To achieve local consistency, we design the temporal variation by imposing an autoregressive model for each time series with coefficients as learnable parameters. Different from previous spatial and temporal regularization schemes, the minimization of temporal variation can better characterize temporal generative mechanisms beyond local smoothness, allowing us to deal with more challenging scenarios such as “blackout” missing. To solve the optimization problem in LATC, we introduce an alternating minimization scheme that estimates the low-rank tensor and autoregressive coefficients iteratively. We conduct extensive numerical experiments on several real-world traffic data sets, and our results demonstrate the effectiveness of LATC in diverse missing scenarios.

Pedestrians are vulnerable road users, and despite their limited representation in traffic events, pedestrian-involved injuries and fatalities are overrepresented in traffic collisions. However, little is known about pedestrian exposure to the risk of collision, especially when compared with the amount of knowledge available for motorized traffic. More data and analysis are therefore required to understand the processes that involve pedestrians in collisions. Collision statistics alone are inadequate for the study of pedestrian–vehicle collisions because of data quantity and quality issues. Surrogate safety measures, as provided by the collection and study of traffic conflicts, were developed as a proactive complementary approach to offer more in-depth safety analysis. However, high costs and reliability issues have inhibited the extensive application of traffic conflict analysis. An automated video analysis system is presented that can (a) detect and track road users in a traffic scene and classify them as pedestrians or motorized road users, (b) identify important events that may lead to collisions, and (c) calculate several severity conflict indicators. The system seeks to classify important events and conflicts automatically but can also be used to summarize large amounts of data that can be further reviewed by safety experts. The functionality of the system is demonstrated on a video data set collected over 2 days at an intersection in downtown Vancouver, British Columbia, Canada. Four conflict indicators are automatically computed for all pedestrian–vehicle events and provide detailed insight into the conflict process. Simple detection rules on the indicators are tested to classify traffic events. This study is unique in its attempt to extract conflict indicators from video sequences in a fully automated way.

Road collisions are a worldwide pandemic that can be addressed through the improvement of existing tools for safety analysis. A refined probabilistic framework is presented for the analysis of road-user interactions. In particular, the identification of potential collision points is used to estimate collision probabilities, and their spatial distribution can be visualized. A probabilistic time to collision is introduced, and interactions are grouped into four categories: head-on, rear-end, side, and parallel. The framework is applied to a large data set of video recordings collected in Kentucky that contains more than 300 severe interactions and collisions. The results demonstrate the usefulness of the approach for studying road-user behavior and mechanisms that may lead to collisions.