Abstract:
Rapid urbanization has made traffic accidents a health and development threat. Accurate urban accident forecasting improves police force pre-allocation and traffic administration and traveler route planning, minimizing injuries and damages.
Off-the-shelf short-term accident forecasting methods model static region-wise correlations with neural networks on hourly and single-step levels. Given the dynamic nature of road networks and expanding urban areas, it is difficult to improve spatiotemporal forecasting as accident records become rarer and long-term future dependencies become more complex.
We propose RiskSeq, a unified framework to predict sparse urban accidents with finer granularities and multiple spatiotemporal steps. We embed region-wise proximity measurements and temporal feature differential operations into a new Differential Time-varying Graph Convolution Network to dynamically capture traffic variations.
By adding contextual factors to a step-wise decoder, a hierarchical sequence learning structure is created. Risk-gather and risk-assign networks learn multi-scale spatial risks to improve risk predictions. Extensive experiments show RiskSeq can improve two datasets by 5–15%.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
