Low-Cost CNN for Automatic Violence Recognition on Embedded System

Due to the increasing number of violence cases, there is a high demand for efficient monitoring systems, however, these systems can be susceptible failure. Therefore, this work proposes analysis and application low-cost Convolutional Neural Networks (CNNs) techniques automatically recognize classify suspicious events. Thus, it possible alert assist process with reduced deployment cost. For purpose, dataset non-violence actions in scenes crowded non-crowded environments was assembled. The mobile CNNs architectures were adapted obtained classification accuracy up 92.05%, low parameters. To demonstrate models’ validity, prototype developed by using an embedded Raspberry Pi platform, able execute model real-time 4 frames-per-second speed. In addition, warning system pre-fight behavior anticipate violent acts, alerting security potential situations.