| The path is the particularity and essence of Chinese classical gardens.In order to explore the inherent law of the ingenious combination of the path space,tourist behavior characteristics can be used as the breakthrough points to carry out quantitative analysis.There are a large number of visitors in landscape area daily,and the paths are complex as well.It is unrealistic to use the traditional method of counting the number of people.In the long run,to repeat visitor detecting and reduce subjectivity as well as randomness of the study,the machine vision technology was applied to the quantitative study of landscape by innovative thinking in the research.Then,basing on the concept of"stagnation",we counted the number of lingering visitors during fixed period of time using image processing and pattern recognition,and analyzed the distribution density of their locations.The Lingering Garden is considered as the research object in this paper.By field investigation and research,we designed and built up the Lingering Garden video monitoring system,to gather the real time data of stationary points.Object segmentation and optimization were carried out by using image pre-processing technology and target detection algorithm.Multi-target tracking came true by comparing of various tracking method.Further more,we designed two adaptable algorithms of counting lingering visitors and locating them on plan,and completed the core mandates.The main research contents can be summarized as follows:(1)Aimed at complex conditions,such as cascading rockery,water fluctuation and wicker swing in the Lingering Garden,we deeply analyzed a variety of foreground extraction algorithms and proposed a background subtraction method based on YUV compensation.By means of doing difference between the current frame and the background U,V channel and then merging,the lost part of foreground is provided after background subtraction.(2)Compared to the current various object tracking technologies,we should choose the most practical tracking method in our subject.On the basis of this,the Matching Counting algorithm was embedded.When calculating whether each target was dwell object or not,we should firstly correspond the storage unit which appeared in last time,transformed time to frame rate and calculated each target resident time.(3)Analysis of proving stagnation point.We deeply studied the various particle filter algorithms based on moving target prediction.On this basis,by combining Kalman filter with lingering counting algorithm,we firstly output the locations of lingering tourists to scene graph.Then,output them to Lingering Garden plane graph by red pixel.we can distinguished whether the place was the stagnation point or not by calculating the density of pixels within the detection region.(4)According to a specific requirement of engineering project,we fleetly designed and established a visitor counting system.The optimal plan was adopted in this system and the updating module was joint,updating the data(5s,6s,8s,10s,12s)at the speed of 5 minutes and outputting in the form of Excel.It was more efficient than getting data by operate software for each residence time.(5)To testify the effectiveness of the algorithm,we stochastically selected 33 stagnation points to count manually and compared data by operating software.The average accuracy was 80.6%.In addition,we also compared the coordinate on a plane graph with the scene location.The average accuracy reached 85.1%. |
PDF Full Text Request