This paper proposes a Genetic Neural Network (GNN) classification method for real-time accident detection at intersections using acoustic signals. Back-Propagation Neural Networks (BPNNs) have been widely used in pattern classification. They have very fast computation speeds that are desirable for realtime detection systems. However, they suffer from the susceptibility of converging to local optimums, which consequently affects the classification accuracy. The proposed GNN uses a Genetic Algorithm (GA) to improve the BPNN global searching ability. The performance of the GNN is compared with that from a widely used Probabilistic Neural Network (PNN) based on detection rate, false alarm rate, and detection time. The test results show that the GNN can achieve performance comparable with the PNN while using much less computation time. The transferability analysis of the GNN also shows that it is robust and can be applied successfully to intersection accident detection based on training data from different sources. The new GNN is computationally inexpensive and highly accurate, and is thus suitable for real-time application.
Application of genetic neural networks in real-time intersection accident detection using acoustic signals
Anwendung künstlicher neuronaler Netze in Echtzeit-Unfallnachweis mittels akustischer Signale
2006
29 Seiten, 6 Bilder, 3 Tabellen, 34 Quellen
Aufsatz (Konferenz)
Datenträger
Englisch
Transportation Research Record | 2006
|Transportation Research Record | 2006
|British Library Conference Proceedings | 2006
|Monitoring system for Intersection detection accident
Europäisches Patentamt | 2016
|