Environment Anomaly Detection In Video Surveillance Via Dynamic Background Modeling

Environmental anomaly detection refers to the detection of people or objects that cause changes in the environment in the monitoring scene.There are many interference factors in anomaly detection under natural environment monitoring,such as wind,light change,water ripple fluctuation and low contrast of image caused by fog.The scene background containing these interference factors is relatively dynamic.The dynamic background brings great challenges to the background modeling.In this thesis,how to accurately and efficiently detect the abnormal changes in the monitoring environment in the dynamic scene is studied.The main work is as follows:1.Research and analysis of the current mainstream background modeling methods,and comparison of the performance of these background modeling methods in different scenarios,and analyze the advantages and disadvantages of each algorithm in detail.2.We analyze the common external environment factors in video monitoring,and discuss the fog and shadow in detail.An improved dark channel defogging method is proposed,which improves the filtering speed of the dark channel defogging algorithm by combining the method of up and down sampling.At the same time,a method of calculating the atmospheric light value in different regions is adopted to make the calculation of the atmospheric light value more accurate.The experimental results show that the improved algorithm not only improves the processing speed of the fog image,but also improves the quality of the fog image.In the process of shadow removal,the foreground moving object including shadow is segmented by using the printing effect of Gaussian mixture model on the illumination change and other factors;then,based on this segmentation,the shadow model of HSV color space is used to detect the shadow in the foreground area,and the contour detection method is used to remove the contour of the illumination edge of the foreground object,so as to get the foreground object without shadow accurately.3.We propose a parameter adaptive Gaussian Mixture Model,which uses the detection results of the previous frame to analyze the current situation of environmental factors in the scene,and uses the foreground dispersion degree to calculate the adaptive parameter threshold.The experimental results show that the improved method is more robust to the noise factors in the natural environment and is suitable for the detection of environmental anomalies in the dynamic background.