A Neuro-Fuzzy Crime Prediction Model Based on Video Analysis

A hybrid neuro-fuzzy model for predicting crime in a wide area such as a town or district in presented. The model is built using what we describe as crime indicator events extracted from simulated wide area surveillance network. The framework principally involves two phases, namely video analysis and crime modeling phases. In video analysis a concept based approach for video event detection is used to detect crime indicator events. Based on the extracted indicators with other related variables, a fuzzy inference system capable of learning is constructed in the second phase. The model is constructed using Violent scene detection (VSD) 2014 dataset and testing is done using UCR-Videoweb dataset. The experimental results show that the proposed method is quite demonstrative and promising.