Road safety is tackled and an intelligent deep learning framework is proposed in this work, which includes outlier detection, vehicle detection, and accident estimation. The road state is first collected, while an intelligent filter, based on SIFT extractor and a Chinese restaurant process is used to remove noise. The extended region-based convolution neural network is then applied to identify the closest vehicles to the given driver. The residual network will benefit from the vehicle detection process to make a binary classification on whether the current road state might cause an accident or not. Finally, we propose a novel optimization model for optimizing hyper-parameters in deep learning methodologies by using evolutionary computation. The proposed solution has been tested using benchmark vehicle detection and accident estimation datasets. The results are very promising and show superiority over many current state-of-the-art solutions in terms of runtime and accuracy, where the proposed solution has more than 5% of improved accident estimation rate compared to the conventional methods.
Hybrid RESNET and Regional Convolution Neural Network Framework for Accident Estimation in Smart Roads
IEEE Transactions on Intelligent Transportation Systems ; 23 , 12 ; 25335-25344
2022-12-01
1462312 byte
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
BASE | 2022
|IEEE | 2019
|Accident prediction models for urban roads
Elsevier | 2001
|Accident prediction models for urban roads
Tema Archiv | 2012
|