Sistema Modular para Detecção e Reconhecimento de Disparos de Armas de Fogo

• Ano de defesa: 2015
• Autor(a) principal: Reis, Clovis Ferreira dos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Informática; Programa de Pós-Graduação em Informática; UFPB
• Palavras-chave em Português: Sons Impulsivos; Armas de fogo; Detecção de sons Impulsivos; Reconhecimento de disparos; Impulsive Sounds; Fire Guns; Impulsive Sound Detection; Gun Shot Recognition; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/tede/9244

The urban violence has been increasing in almost Brazilian state and in order to face this threat, new technological tools are required by the police authorities in order to support their decisions on how and when the few available resources should be employed to combat criminality. In this context, this work presents an embedded computational tool that is suitable for detecting gun-shots automatically. To provide the necessary knowledge to understand the work, a brief description about impulsive sounds, re guns and the gun-shot characteristics are initially presented. Latter, a system based on modules is proposed to detect and recognize impulsive sound, which are characteristics of gun-shots. However, since the system contain several modules in this work we have focus only on two of them: the module for detecting impulsive sounds and the module for distinguish a gun-shot from any other impulsive sound. For the impulsive detection module, three well-known algorithms were analyzed on the same condition: the fourth derivative of the Root Median Square (RMS), the Conditional Median Filter (CMF) and the Variance Method (VM). The algorithms were tested based on four measured performance parameters: accuracy, precision, sensibility and speci city. And in order to determine the most e cient algorithm for detecting impulsive sounds, a cadence test with impulsive sounds, without or with additional noise (constant or increasing) was performed. After this analysis, the parameters employed on the CMF and VM method were tested in a wide range of con gurations to verify any possibility of optimization. Once this optimal method was determined, the classi cation module to recognize gun-shots started to be implemented. For this, two distinguish methods were compared, one based on the signal wrapped over the time and the other based on most relevant frequencies obtained from the Fourier transform. From the comparison between the two methods it was observed that the wrapped method provided 54% of accuracy in the classi cation of impulsive sounds, while with the frequency analysis this value was 72%.