Background Modeling And Object Counting From Video Streams Based On Improved Adaptive Gaussian Mixture Model

In daily life, the camera is often installed in the parking lot, traffic junctions, airports, Banks andother places to monitor events.Usually we should configure person to observe the video and captureinformation, due to the need for continued concern, this is a tedious and time-consuming work.If wecan develop an automatic system to analyze video streaming and get the information, then makecorresponding action, we can save a lot of manpower and material resources.So it has very importantpractical significance and economic value.In this paper, we make in-depth research onbackground modeling method based on GMM fromthe point of image process based on analysis of existing motion detection and trackingmethods.Simultaneously,we use the KLTmethod which iscurrently considered mosteffectiveontracking object. The main contents of the present paper include:1.The improvement ofGaussian mixture model algorithm.First of all, We analyze the defects ofthe existing background segmentation algorithm; Then, according to these defects, we put forwardsome improvements, they are mainly divided into the following five aspects:1) background modelingbased onblock;2) background model criterion;3) initialization of model parameter;4) model studyand the background updating;5) light mutation detection.2. Background modeling. In order to verify the influence of different environment on theimproved algorithm, we do experiment in different scenarios. Experiment is divided into fiveaspects:1)Analysisthe algorithm segmentation performance in outdoor environment;2)Compare thesegmentation results in indoor and outdoor environment;3) Validate the influence of speed;4)Thedetection of the algorithm of light mutation;5)The segmentation performance of our algorithm in acomplex scene.Experiments prove that our algorithm is better than the original algorithm.3.Object counting. We put forward an object counting system based on the improved backgroundsegmentation algorithm.This system uses KLT tracking algorithm. Firstly, we use the improvedbackground modeling method to detect object; Secondly, we use the KLT algorithm to detect featurepoints ofthe segmented pedestrians and track them; Then, cluster the finished tracking trajectory;Finally, output the clustering results and get the number of passengers. Oursystem obtain highreliability and robustness.This paper makes in-depth study for moving object detection and tracking algorithm undercomplex scene and proposes anovel, effective and feasibleschemefor the target segmentationandtarget tracking.