Crowd abnormal detection using artificial bacteria colony and Kohonen's neural network

This paper presents a new method for detecting abnormalities in crowded scenes using Artificial Bacteria Colony. The proposed method uses a metaheuristic inspired by the behavior of colony formation of bacteria. Artificial Bacteria Colony are used to optimize the search for moving areas on image. The detection method using the algorithm of Artificial Bacteria Colony is robust exhibiting an ability to adapt quickly to any scenario and the overall result is not impacted by the noise from videos. The bacteria population, the food stock and the centroid of the colonies are used as data for training a Kohonen's neural network. After training, the network is able to detect specific events by the similarity of data. The experiments were performed using the public dataset UMN. The results show that the proposed scheme is similar to state-of-the-art algorithms for detecting abnormalities nn the behavior pattern of people in crowds.