Study On Algorithms Of Moving Object Detection Based On Local Gray Distribution

With the development of the network communication technology, microelectronics,artificial intelligence and computer vision technology, people pay more attention to theresearch and application of the intelligent security systems. The moving object detection is thefirst step of the intelligent security system and is a very important step, it processes videoimage sequences, extracts the moving objects, provides data to support the following worksuch as moving object tracking, behavior analysis. Moving object detection technology hasbecome one of research topics in computer vision, and it has a high value in scientificresearch and engineering applications, researchers proposed a variety of motion detectionalgorithms, but through practices we found monitoring the scene is not easy, such as branchesand leaves swinging, water fluctuations and light transform, it will affect the moving targetdetection. Therefore, how to monitor the scene in a complex scene is a more difficult problem.Back ground subtraction method is one of the most widely used methods for the detectionof moving targets, reference to the methods of the Gaussian mixture model and the LBPmodel, this paper studies a new background subtraction method - algorithms of movingobjects detection based on local gray distribution. It separates the current image into multipleregions with a certain size, and conducted the statistical of the histogram of the graydistribution which operates spatial weighted by Epanechnikov kernel function in the region,we call it the histogram of distance and gray distribution, and we use the normalized distancevalue from each pixel to the center of this region as the variables of the Epanechnikov kernelfunction, this model describes the characteristics of the region, in this way, the characteristicmodel of this region has the edge of the robustness. This paper divides the feature models ofthis region into foreground feature models and background feature models, and we use thefeature models of this region in multiple times to be the multiple models of the backgroundmodel of this region, and give each model a certain weights, then we use the improvedhistogram intersection method to match the characteristics model of the new image with theeach model of the same region in the background model, if successful, this region would bejudged as the background region, otherwise, judged the region as the foreground region,finally, we update the background model using different methods depending on the matchresults. We compare the prospect targets extraction using this paper's method and the mixedGaussian model in different monitoring scene, the experimental results show that this paper'smethod exclude the changes of the objects into and out of the background in the scene and adapt to the changes of the light gradient and the light mutations in the background, and thecontours of the extracted prospect targets are clear and complete, and the extraction ofmultiple moving targets is also very good. Time complexity of this paper's method is low,meet the real-time requirements. In addition, we apply this method to the security system, thefalse alarm rate and false negative rates are very low, and detects effectively.