Algorithm On Fence Climbing Detection For Perimeter Video Surveillance

With the installation of a large number of cameras in public places,the demand for intelligent video surveillance system is more and more urgent.Fence climbing is a common intrusion behavior.While,in this direction,the theoretical research is still in its infancy.Based on the behavior analysis algorithm,some studies extract the characteristics of human skeletal nodes and determine whether some fence climbing behavior has occurred in the video or not.All these methods haven't taken into account some certain practical conditions containing complex occlusion and multiple moving objects cases,making it difficult to achieve the satisfying result given in the papers.To tackle the problems discussed above,this paper is focused on providing a fence climbing detection algorithm with a steady effect in the practical environment.Based on the previous algorithm frameworks,this paper has added the detection and tracking modules,implementing the algorithm process of object detection-human tracking-trajectory analysis.In the object detection module,the detection of suspicious persons was implemented by using two level classifiers composed of the Bayesian classifier and Adaboost classifier.While in the human tracking module,the suspicious persons were tracked by integrating the KLT algorithm,the moving foreground and the prediction results of Kalman filter.This module can getting a steady motion trajectory.Besides,in the trajectory analysis module,having abandoned the skeleton nodes features that can't be calculated stably in the realistic environment,this paper has adopted the stable trajectory characteristics to analyze human motions and achieve the final result.In this paper,many test videos were recorded on different occasions to test the accuracy of the algorithm.The algorithm has an accuracy over 93.3% in the testing videos without any false positives.With the processing time is only 10 ms for each frame,the algorithm was also able to cope with the various complex situations in the practical environment.Results showed that this algorithm has a great accuracy,robustness and real time performance.Compared with the previous methods,this approach can better adapt to the complicated environmental conditions in practical applications.