Research On Moving Object Detection Methods In Atmospheric Turbulence Environment

The moving object detection extracts the target of interest from videos or image sequences.It facilitates further research on object tracking,classification and understanding of behavior.It is very important and has become one of the hot research directions because of the popularity of monitoring equipment.Images acquired from a long distance are easily influenced by the atmospheric turbulence,showing a different state from the real scene.Atmospheric turbulence is a kind of turbulent motion induced by temperature,humidity,pressure,wind speed,etc.It makes the atmosphere differ in density and the refractive index,thus affecting the light transmission.Images are affected by atmospheric turbulence to various degrees,mainly for the changes of intensity and random drift.Traditional moving object detection methods are hard to tackle this complex situation.This paper conducts research on moving object detection in atmospheric turbulence environment.The paper introduces some traditional moving object detection methods and analyses their strengths and weaknesses.Then it studies the impact of atmospheric turbulence on images.In view of the different effects of atmospheric turbulence on image sequences in different regions,an adaptive moving object detection method based on multi-regional modeling and multi-hierarchy decision is proposed.The method divides an image into four regions,namely,the flat region in the background,the edge region in the background,the abrupt changing pixels and the moving object.It models these regions at three levels.At the first level,the method uses parametric and non-parametric methods to preliminarily model the background.The background image is divided into a flat region and an edge region.Different models are applied by studying the properties of the two regions.The parametric method uses Gaussian distribution and double Gaussian distribution to model the flat region and the edge region in the background respectively.The non-parametric method which is based on Vi Be uses different thresholds for the two regions to obtain a preliminary foreground image.At the second level,the paper studies an adaptive thresholding model to identify the false alarm.Due to the intensity abrupt changing pixels,the preliminary foreground image contains some false alarm regions.The adaptive threshlding model integrates the global standard deviation information and the local neighborhood pixel information,which can effectively eliminate the abrupt changing pixels.Based on this model,the paper studies a ghost elimination method which can eliminate the ghost rapidly.It increases the threshold exponentially by counting the number of pixels continuously being detected as foreground.At the third level,the method filters the foreground image by morphological operations and obtains the moving object by the connected component restriction.Comparison with other methods shows that the method performs well under different situations such as variation on different turbulence strength and different numbers of objects and moving directions.