Proceedings of The 11th International Conference on Research in Engineering, Science & Technology
Structure-Based Pattern Recognition to Aid Identification and Criminal Proceedings of Disclosure in Collective in Belo Horizonte
Welton Pereira Da Silva Neves, Gabriel Dias De Lima Novais, Luiz Melk de Carvalho, Diva de Souza e Silva Rodrigues, Flávio Henrique Batista de Souza
During the criminal action of theft and robbery of public transportation, sudden movements and the lack of real-time information about the event’s occurrence can cost the patrimony or even the lives of innocent people. This paper demonstrates a proposed structure, based on Multilayer Perceptron (MLP) and Support Vector Machines (SVM), Internet of Things (IoT), Microcomputers and Cloud Computing; to assist the monitoring of public transport in the city of Belo Horizonte (3 million inhabitants) and help on the recognition of the behavior of a criminal event. The proposed structure captures conversations on the bus and, when identifying terms from an assault, considers the bus as in a risk situation, without the need for exposure of the victims during the occurrence. A comparative analysis was carried out based on the Area Under the Curve (AUC), of the pattern recognition structures, where it was used: a database with a 57-word dictionary; six types of MLP, with variations of: neurons in the hidden layer, times and the test and validation set (constituting 108 tests and an average execution time of 42.36 seconds); SVM with K-Fold Cross Validation, with three types of Kernels (averaging 8.51 seconds). An AUC of 0.948 was obtained via SVM. This structure has an implantation architecture based on Microphones, Raspberry Pi 3 Model B, and Google Voice API for capturing and processing speech during the collective journey. The system administration consolidation uses a Cloud Computing platform and it was estimated that each unit could cost between 400 to 600 Brazilian reais.
Keywords: Pattern Recognition; Collective Transports; Criminal Occurrences; Multilayer Perceptrons; Support Vector Machines.