A Robust Moving Object Detection In Multi-Scenario For Video Automatic Analysis

Modern video surveillance benefits greatly from advanced wireless imaging sensors and cloud data storage;thus,considerable video footage can be generated every second.Surveillance videos have thus become one of the largest sources of unstructured data.Using such multiscenario videos for detecting moving objects is a challenging task for users of conventional moving object detection methods.Moving object detection,as the key role of automatic video analysis,affects the accuracy of video automatic analysis technology due to its accuracy and robustness.Based on the above problems,we proposed novel moving-objects-detection models in multiscenario.The main works of this thesis include:1)Aiming at the problem of detection accuracy and robustness of multiple scenarios detection(i.e.,moving object detection in single scenario at one time)under multiple scenarios,this thesis proposes a Low-Rank representation with Contextual Regularization(LRC)model.The LRC model uses not only low-rank constraints to represent the background model in the video,but also consider the moving objects with context regularization.The low-rank representation ensures that the LRC model be more effective than the traditional detection methods in multiple scenarios without the hassle of parameters in a single scenario.2)In order to solve the problem of robust moving object detection of multi-scenarios in series,Sparse and Low-Rank representation with Contextual Regularization(SLRC)based on LRC model is further proposed in our thesis.After the LRC transforms the moving object detection problem into a continuous outliers detection problem with low-rank representation,a sparse coefficient learning method is further used to construct a dedicated background for each single scene of multiple scenarios.The SLRC model successfully transforms the video detection problem of single multi-scene detection into a multiple single-scene detection problem.3)This thesis further employ LRC model to the application of traffic congestion detection.Fusion of visual features and the detection of moving objects then are used to determine the state of the road.The model firstly calculates the road occupancy rate and traffic flow based upon LRC detection.Then,based on the virtual detector,the GLCM-based density analysis and the optical flow-based velocity analysis are processed on the input image.Finally,the above multi-dimensional visual features are combined to detect the state of the traffic road.Experiments and analyzes indicated that the moving object detection models proposed in this thesis can achieve effective and robust practical performance both qualitatively and quantitatively.And through the application and experiment in the traffic road congestion detection,it also shows that it plays an important role in intelligence traffic system.