Detection of static groups and crowds gathered in open spaces by texture classification Marco Manfredi, Roberto Vezzani, Simone Calderara, Rita Cucchiara DIEF - University of Modena and Reggio Emilia Tel. +39 0592056270 Fax. +39 0592056129 {marco.manfredi, roberto.vezzani, simone.calderara, rita.cucchiara}@unimore.it Abstract A surveillance system specifically developed to manage crowded scenes is described in this paper. In particular we focused on static crowds, composed by groups of people gathered and stayed in the same place for a while. The detection and spatial localization of static crowd situations is performed by means of a One Class Support Vector Machine, working on texture features extracted at patch level. Spatial regions containing crowds are identified and filtered using motion information to prevent noise and false alarms due to moving flows of people. By means of one class classification and inner texture descriptors, we are able to obtain, from a single training set, a sufficiently general crowd model that can be used for all the scenarios that shares a similar viewpoint. Tests on public datasets and real setups validate the proposed system. Keywords: Crowd Detection, Surveillance, One Class SVM 1. Introduction Streets of our cities are day by days more crowded, impacting our well-being in comfortable environments, affecting our sense of safety and also creating serious problems of security. Studies on crowds and individu- als in a crowd are critical for surveillance and real-time proactive control of safe and smart cities, since crowds could cause or be caused by violent events [1]. Several times in the past organized groups of hooligans met up in public areas, scheduled a plan and, armed, they went to the stadium just to look for a brawl; groups of riot- ing people met each other to create an untamable flow of violence. Conversely, violent events, accidents and unexpected situations can create crowds (for instance at the exit of undergrounds, at the gates of building, etc.) which, in turn, induce additional public safety prob- lems. The pioneering definitions of crowd discuss the si- multaneous presence of densely distributed high num- ber of individuals. In this context studies from Le Bon and Freud in early 20th century treat the crowd as an emotional mass with a intuitive loss of consciousness from single individuals [2, 3]. We refer to this type of crowd as dense crowd. Theory of dense crowd of Gus- tave Le Bon is based on the “contagion phenomenon”, in which the crowd goal and formation mechanisms are emerging from a global crowd will, where individuality is lost. Recent studies observe the collective behavior of crowd under a different perspective. In Turner and Kil- lian [4] observation, the crowd phenomenon can emerge even from small gatherings of people diffuse in a large area, authors refer to this case as the diffuse crowd case. In diffuse crowd, individuality is more observable than in dense crowd. For this reason, the “crowd will” towards a common goal is often not directly verifiable. The behavioral pattern emerging in this situation fol- lows the “convergence theory”: the individual will pre- vails over the collective, resulting on a crowd formation due to the gathering of people sharing a common goal. Under these premises automatic tools for crowd anal- ysis should take into account the type of the phe- nomenon object of the study. The solution proposed in this paper focuses on the surveillance of open environments where the gathering of groups of people generates a diffuse crowd. The pur- pose of the system is to aid the security officers to timely identify security alerts and eventually acquire the iden- tities of the authors of the crime [5]. In the aforemen- tioned scenario a single or a set of Pan-Tilt-Zoom cam- Preprint submitted to Pattern Recognition Letters April 28, 2014