Research And Design Of Binocular Stereo Vision Based People Counting System

At present,the bus routes in many cities are not satisfying.Especially in the major city,the bus routes become extremely complex.Since lacking of enough passenger data,it is very difficult for the bus company to arrange the bus routes properly.This problem causes an extraordinary crowded environment on some particular buses,and seriously impacted the passenger experience.Therefore,the bus company needs amounts of the passenger data to design the bus routes reasonably and make the bus dispatching more intelligent.From this starting point,this thesis proposes a people counting system that can be used in public transport system such as bus and subway.We utilize the binocular stereo vision technology to collect the passenger data automatically,the main work and achievements are summarized as follows:(1)Firstly,we do some works on the stereo matching algorithm.The Census transform is very useful in stereo matching algorithms.But the traditional Census transform costs too much computation,and its accuracy and robustness are not satisfying.In this thesis,a modified semi-global matching(SGM)algorithm based on the fast Census transform is proposed to improve the quality of the range map and decrease the computation.In addition,we modify the disparity refinement to enhance the quality of the estimated range map.Our experiments show that the proposed algorithm performs better than the original SGM algorithm in terms of the accuracy and the anti-noise capability.(2)Secondly,we work on the object detection algorithm based on the range map.The limited precision of the range map,the different heights of the passengers and the crowded environment on the bus,make the stereo vision based multi-object detection become very difficult.In this thesis,we analyze and compare several existing object detection methods,and propose an object detection method based on the connected components analysis.The method can detect multiple objects effectively,and is robust to the different heights of the passengers.Compared to the original method,our method decreases the computation greatly,and can be ported much more easily.(3)In order to collect the passenger data accurately,we study on the object tracking and counting algorithm.This thesis proposes a multi-feature based object tracking algorithm,which overcomes the problem of the original algorithm on the fusion of different features.The proposed algorithm perform very well for the multi-object tracking,because it can make adjustments adaptively,which can ensure the correct tracking for multiple objects.In addition,this thesis presents some judgement strategies to determine whether the passenger is entering,or exiting.(4)Finally,we port the people counting algorithm to an embedded platform.In this part,we not only describe how to build a binocular stereo vision system for a particular environment,but also parallelize the main algorithm modules by OpenCL.