Abnormal Behavior Detection Of Passengers In Elevator

Elevator is a place likely to occur failing-down,violation,robbery and crowd panic for the space of the elevator is small and confined.It's of great realistic significance for safeguarding people's security to prevent the events mentioned above from happening.Abnormal behavior detection in elevator has been researched deeply in this paper and some useful results are achieved.The main work of this paper is summarized as follow:1.A complete system based on image detection for abnormal behavior detection in elevator is designed and realized.The system obtains the human target by background subtraction first of all,and then estimates the numbers of passengers through counting pixels of the region of human object.2.Different detection models and algorithms are designed according to the num-bers of passengers.The system checks out whether someone is failed down or stay down through processing the projections of interest region when there is only one passenger in elevator.Violence behavior is detected when there are two passengers in elevator.The system calculates motion history map of passengers to obtain accumulated motion in space and time and energy function is formulated based on motion history map.Finally the system judges whether the violent behavior exists or not based the value of energy function.When there are three or more passengers in elevator,the system checks out whether there is crowed panic behavior.The system calculates motion energy map of passengers and the histograms of energy distribution are obtained by projecting the mo-tion energy map to horizontal direction and vertical direction.Crowd panic behavior is judged according to the characteristics of the space distribution of the histogram.3.To verify the result of the methods proposed in this paper about abnormal behav-ior detection of passengers in elevator,a dataset about passengers behaviors including different abnormal behaviors in elevator is collected and provided.The dataset has 34 videos of single passengers,28 videos of two passengers and 26 videos of crowed.The methods proposed is this paper are tested on the dataset and the experiment shows that the abnormal behavior detection algorithms came up in this paper are concise and able to work reliably in real time.