Research And Implementation Of People Counting In High Crowd Scenes Of Tongji Bridge

Tongji bridge is a long history building of Foshan. People walk across the bridgeto make a wish in Lantern festival day in Foshan. Tongji bridge and the surroundingarea has a high crowd in that day. Automatic monitoring the number of people in thearea can provide important information for the infrastructure construction, publictransport adjustment, crowd accident and people travel.In summary, there is great practical significance to the research of intelligentpeople counting in Tongji bridge. In this paper, we proposed a new method ofpeolple counting that makes a step forward in resolving the the Tongji Bridgeintelligent counting problems by studying the existing method. The main work of thispaper is as follows:1. The thesis does research on people counting method base on targetdetection, analysis the characteristic of Tongji bridge surveillance, and study threeclassic head detection method: haar feature detection, integral channel featuredetection, and HOG feature detection, discuss the feasibility of counting in TongjiBridge surveillance video using three of them.2. The thesis analysis the reason that generation high false positives of usingHOG feature detection in Tongji Bridge surveillance video. Proposed a detectionmethod of combination of HOG features and color histogram features. This methoddetermine the target through joint SVM calculations of both features, can eliminatemost of the false positives HOG feature detection generated. In the problem ofcalculating people through the surveillance region, we design a tracking and countingmethod using region matching base on optical flow. This method combine proximitydetection target in adjacent frames to complete tracking and counting, and find anew match by updating position continuously using historical speed when target loss.This method can provide some fault tolerance when the target loss because of occlusion, and reaches a relatively good accuracy in the complex environmentsurveillance video of Tongji Bridge.3. Experimental Simulation and Analysis. Firstly, the thesis provides theparameters of performance evaluation, including Mean Absolute Error, Mean Miss,Mean False, etc. Secondly, the thesis tests the detection method combine HOGfeature and color histogram feature using two kinds of density surveillance video inTongji bridge, and compare the result with HOG feature detection. Thirdly, the thesistests the region matching method in counting people flowmeter, and discuss theresult in detail. Finally, the thesis analysis important parameters through experiment,and give the best selection at last.