Research And Implementation Of Indoor People Counting Method Based On Video

Video surveillance equipment has been widely used in schools,shopping malls,banks,stations etc...Managing human density by detecting the number of people is meaningful in these occasion.The research of people counting in indoor spaces based on video get the information of human flow through analyzing and processing video images automatically.It is an important tendency of computer vision.And it has significant research value and broad application prospects.This paper analyzes the domestic and foreign research status quo such as template matching,color model and classifier detection.And research also does analysis and comparison about the method of existed human detection and statistics.Paper get a favorable method and technology of people counting method in indoor spaces.The main work is as follows.1.The principle and implementation process of head detection based on AdaBoost(Adaptive Boosting)algorithm.This section detailedly analyzes the computing way of eigenvalues and the training process of classifier and cascade classifier in AdaBoost algorithm.Head detection and statistics use double iterative detection strategy to come out.2.Head detection combined edge feature of head-and-shoulder and improved AdaBoost algorithm.Traditional AdaBoost algorithm has some shortcoming such as large numbers of training samples,low speed and low accuracy.Head detection comes more effective in speed and accuracy by improving the training process of classifier and combining it with the edge feature of head-and-shoulder.3.Based on the result of head detection,we achieved people counting by integrating foreground checking.Taking the result of head detection as human number is acceptable when people is not too much.But it is not suitable when human become dense.With the increase of human density,the detection rate of this method reduced immensely.The way of foreground checking could correct detection result and then achieve people counting.4.Designed system complete testing and analyzing of function and performance.We test a large number of experiments based on classroom surveillance video data.The results show that our method is able to detect the number of people in video image accurately and quickly.Regardless of the change of crowd mobility and crowd density,this method still has better detection performance.The research work get the method of people counting by combining the edge feature of head-and-shoulder,improved AdaBoost and foreground checking.The designed system verified that it was effective,precise and real-time.