Vision based surveillance system

Due to the numerous amounts of surveillance cameras available, security guards seem to be ubiquitously watching over. However, the number of existing cameras exceeds the number of humans to monitor them and the supervision of all the sensors´ output is costly. Thus, video footage from cameras is most often only used as a forensic tool. This suggests the need of an intelligent video surveillance system providing continuous 24-hour monitoring, replacing the traditional ineffective systems. This paper presents an automated vision based surveillance system which is capable to detect and track humans and vehicles from a video footage. Simulation results have shown that the Object Classification module manages to achieve an accuracy of 97.31% and 97.14% for the person and vehicle classification respectively. Furthermore, the system manages to successfully track the objects 97% of the time under no occlusion and 94.14% in presence of occlusion.