Research On Classification Algorithms Of Target Images In Static Background

Image number statistics belongs to the research field of pattern recognition.It refers to the number of people in static images or dynamic videos scene which count by computers or other computing devices.At present,there are relatively few people's statistics on static images.In this paper,a static image covnting algorithm based on machine learning image classification technology is proposed,taking the images collected by a school classroom monitoring system as the research object.Firstly,classical image feature extraction and classification algorithm are introduced.Image feature is an algorithm used to describe the detection of target features by image processing technology.This paper mainly introduces two feature extraction algorithms,HOG and Haar.Classification is a supervised learning method.The parameters of classification model can be obtained by training the specified data of the class label,and then the unknown class data can be recognized.In this paper,three classification algorithms,which are Support Vector Machine,Bayesian and Adaboost,are introduced.Then,a statistic algorithm based on background region detection is proposed.This algorithm takes the seat back in the classroom image as the target detection area.By extracting the HOG features of the area image,a naive Bayesian classification algorithm is designed to judge the occupied state of the seat area,and compared with the SVM classifier in MATLAB.Experiments show that the Bayesian classifier designed in this paper has higher classification accuracy.Finally,a static image counting algorithm based on foreground head detection is proposed.The algorithm uses multi-scale sliding window method,designs the color detection,and trains Adaboost and SVM cascade classifiers for human head target recognition.Experiments show that the algorithm has the advantages of better real-time performance and better accuracy.The experimental results based on building 5 sample libraries show that the proposed detection algorithm based on background region and foreground human head can achieve the number statistics and determine the location of people where are in static images.