Spatio-Temporal Image Processing for Predictive Modeling of Pedestrian Intent and Trajectories in Urban Traffic Environments: A Comprehensive Review

Among the problems that need solutions in the urban transportation domain today, pedestrian safety is major issue, especially considering the increased use of autonomous vehicles. Today's vision systems operate mostly based on the detection and tracking of pedestrians and are only able to react to certain movements, making them unsuitable for safety-related applications. To resolve this issue, it is more necessary now than ever to have approaches that will allow predicting the behavior of pedestrians even before it takes place. In this paper, conduct a comprehensive review of the spatio-temporal image processing techniques for modeling pedestrians' intent and predicting their trajectory. The special emphasis is given to the importance of contextual information, such as scene structure and interactions between road users, for improving the reliability of the predictions. Further, probabilistic modelling approaches are explained to accommodate the uncertainty and multi-modal nature of human motion. An overview on the existing methods is given through comparison with the widely accepted evaluation criteria, showing recent advances in the prediction accuracy and robustness. Despite these advances, issues of interpretability, data scarcity, applying the concepts in a wide range of settings, and the requirement to implement it in real-time remain. Finally, the paper discusses important research directions geared towards the creation of more reliable, scalable and context-aware predictive systems for future intelligent transportation systems.