DETECTION OF SUSPICIOUS ACTIVITY IN VIDEO SURVELLANCE

Abstract

The video surveillance systems have gained popularity since last few decades because of their use in the detection of unusual activities, surveillance, patrolling, and other scientific and engineering problems. Activity detection is an important component of video surveillance and involves tasks like recognition of humans, their activities with respect to their surroundings and the further analysis for any abnormality or suspicious behavior. This recognition can be done either manually or automatically with the help of computers. Though it is very easy for a human to analyze the video for suspicious activities, and this is the way which is in widespread use, the other way can be to do it automatically. Autonomous video surveillance requires automatic processing of video sequences. This work, therefore, proposes the approach to do the surveillance automatically. The detailed approach along with its advantages over other approaches has been discussed at length. The various constraints that have been taken into account are also elaborated. The design of the system takes input from the video frames taken at the place where we provide surveillance. The system does both the low-level processing, like motion detection and tracking, and also performs high level decision making jobs like unusual activity detection. This work, therefore, aims to translate the low-level input into a high-level semantically meaningful activity description. The three major components of the work include moving object detection, tracking and unusual activity detection. The approach in this dissertation is substantiated by taking two unusual activities, first is abandoning of bag by a person and the second is carrying of bag by a person. Only a single person is involved and outdoor background and static background is taken. The analysis is made on offline videos and no real-time detection or analysis is made. The development is done in C++ using OpenCV library on Linux platform.