Realistic Video Analytics Implementation on a Heterogeneous Infrastructure

In real time safety administration requires exact pedestrian crash risk evaluations at various times. This work introduces real-time vehicle-pedestrian collision risk modelling for signalized crossings. Bayesian Generalized Extreme Value modelling assesses post-encroachment traffic conflict accident probability in real time. Pedestrian conflicts during the cycles of traffic signals are identified by the use of block maxima sampling using a Generalized Extreme Value distribution. Bayesian analysis handles signal cycle crash risk and traffic severe heterogeneity using various signal-level parameters. Systematic a covariate extractor from streamed video path, transportation conflict, and signal timing databases yields signal cycle-level covariates to explain cycle-dependent accident risk. Study crash data is calculated correctly. The fitted generalized extreme value distribution forecasts signal cycle-level collapses. The real-time collision risk model shows pedestrian accident risk at signalized crossings all day. This study shows that the proposed real-time system can measure signal cycle-level vehicle-pedestrian accident risk, allowing proactive safety management and pedestrian risk reduction.