Research On The Statistic Method Of The Number Of Visitors Based On The "Stagnation Points" In Lingering Garden

Chinese classical garden is extensive and profound.In order to explore the inherent characteristics of the garden,the pedestrians on its path need to be counted.The number of daily tourists in the garden is huge,and their distribution situation is complex in the patlh.The traditional artificial counting method not only consumes a lot of manpower and material resources but also has low accuracy of data.In order to efficiently and accurately calculate the number of tourists in the station,this project introduces computer vision technology and corresponding machine learning algorithm to carry out quantitative statistics of tourists in the park.The subject of this study which is mainly aimed at the group tourists in the stagnation point is Suzhou Lingering Garden.Pretreat collected video,research team identification algorithm,and output team tourist pictures.The characteristics of team visitors are extracted and the number of team tourists is predicted by the traditional SVR algorithm.This paper also studies a variety of algorithms based on deep learning.On this basis,the paper builds a statistical system of team visitor numbers,develops special software,and realizes the statistics of the number of tourists in the group.The main research contents of this paper include the following points:(1)The pre-acquisition video is classified and processed,and the current practical foreground extraction algorithm is analyzed and run.And the Gaussian Mixture Model is selected and improved by comparison of several algorithms.By optimizing the pre-area ratio parameter,the team's tourist identification algorithm is improved.Then through the perfect team recognition algorithm to complete the collection of team tourist pictures.(2)Analyze the group tourists and extract the main characteristics of the team visitors.Through the traditional SVR algorithm,the number of team tourists is counted and the model performance is evaluated.(3)The current deep learning algorithm is studied and the network structure commonly used in deep learning is introduced.And the target detection algorithm based on Region Proposal is systematically studied,the results of the current mainstream Faster R-CNN and R-FCN target detection algorithm are emphasized.Then,two main algorithms are implemented and analyzed by data,and the target detection algorithm is selected for this research.Finally,by selecting the target detection algorithm and combining witih the human head model,a statistical system for the number of visitors is constructed.(4)TensorFlow depth learning framework and PyQt library were utilized in Visual Studio Code to develop team visitor statistics software.The software system can be implemented by setting parameters,displaying images folder,and viewing the processed images.It also can realize the batch processing of team tourist pictures and output the statistics to excel.(5)The accuracy of the test software is analyzed by the result of the data.And the influence of the station space type,season and other factors on the data accuracy is investigated.The accuracy of the data is calculated by comparing the manual count with the software count.The average accuracy of the data is 90.63%.This paper took the factors that affect the accuracy of the accuracy of the data into consideration for the future model improvement work.