Research And Application Of Video Target Detection Algorithm Based On GMM

In recent years,with the rapid development of computer technology and the Internet,computer technology has provided good technical support for the creation of an intelligent society.Computer-related technologies are widely used in various industries in the society.Among them,the target detection technology in the field of computer vision has become a key technology in the field of surveillance,which is widely used in medical,traffic speed measurement,military strike and other fields.This article conducts in-depth research on the background modeling method,and proposes two improved GMM(Gaussian Mixture Model)target detection algorithms.The main content of the research is divided into the following parts:(1)In the process of target detection,GMM is easily disturbed by factors such as lights,target color similar to background color,target shadow and shooting height.In response to the above problems,a GMM algorithm combining improved HED(Holistically-Nested Edge Detection)network and Otsu double threshold segmentation is proposed.First,the improved model performs dual threshold segmentation on the background,noise,and foreground targets of the video frame,and selects the number of Gaussian models reasonably.Secondly,the HED network is used to perform edge detection on the input picture.The edge result of the HED network detection and the GMM detection result of the double threshold segmentation are AND operation to obtain the final target detection result.(2)The GMM algorithm uses a fixed number of models in target detection to describe the state of pixels and a fixed learning rate to update the background.In view of the above deficiencies of the GMM algorithm,a GMM algorithm that adaptively selects the model number and learning rate to adapt to changes in the environment is proposed.Through fuzzy theory,the video is divided into three fuzzy subsets,then the fuzzy entropy of each part is calculated,and finally the number of models required is determined according to the maximum value of the video's fuzzy entropy.Introduce the correlation of the video frame,measure the correlation between the reference frame and the detection frame,compare the background change factor and the background change coefficient to determine different learning rates.(3)In order to strengthen the protection and effectiveness of Qilian Mountain environment,the improved algorithm is applied to Qilian Mountain environmental protection,mainly to detect Qilian Mountain fire,harmful animals,poaching,etc.Through experimental verification of the performance of the algorithm in(1),it is found that the detection rate of the improved algorithm is higher,the detection contour is more complete when the target is smaller,and the detection effect is better.Experiments verify that the algorithm in(2)can eliminate the influence of noise,effectively save the number of Gaussian models,and adapt to environmental changes.The detection accuracy is improved,and the detection time is reduced.The target detection algorithm is applied to Qilian Mountain environmental protection,which can effectively protect Qilian Mountain ecological environment and forest area safety.