An interactive framework for abandoned and removed object detection in video

This paper describes an interactive application that employs a modular approach to solve the problem of detecting abandoned and removed objects in a video stream in real time. The system breaks down this difficult task into a series of simpler sub-tasks and solves each through a separate module. Since different modules are independent of each other, each may utilize several different methods for solving the corresponding subproblem. Accordingly, several existing methods have been implemented for each module with the possibility of adding new methods with minimum system reprogramming. Additionally, by switching between these methods in real time, the user can observe and compare the performance and accuracy of different methods and find the optimal combination of methods for a particular scenario. Lastly, the user can also interact with the system to provide useful feedback and help improve its performance when it is not up to the mark. This enables the system to take advantage of the far superior human vision processing capabilities in very complex scenarios while still being completely automated most of the time.