Elevator-Waiting People Counting System Based On Computer Vision Detection

Computer vision detection system is one of important high-techapplication fields and front topic in measurement technique domain. It spansmany subjects including computer science, pattern analysis, measurementtechnology, digital image processing and artificial intelligence etc. Computervision detection system can automatically analyze the sequence of images bythe methods of computer vision and video analysis. The system can recognizeobjects, track and detect the position of the objects in a complex environmentin real-time. Furthermore, the system can analysis the image or video andhave its own comprehension and give the accurate measurement results.With the rapid development of the economy and the citilization, thesolutions for vertical traffic problems have been more and more importantbecause the buildings become higher and higher. In order to solve thecomplex vertical traffic problems in the high buildings, efficient elevators-control-system service is a key issue beside the elevator speed improvement.If we can get the accurate number of elevator-waiting people before they gointo the elevators and then send more elevators to the floors with huge crowd,the efficiency of the elevators would be improved to a great extent. Anelevator-waiting people counting system based on computer vision detectionis presented in this article to provide the accurate and real-time peoplenumber using the advantage of computer vision detection technology.This article has analyzed such a particular application in details; take thedifferences of every elevator waiting hall, illumination, target objects andpositions of all the waiting people into consideration. After that, a efficientand robust elevator waiting people counting system is presented, including awide angle hemispherical camera with adjustable focus, image capture card based on H.264encoding, soft decoder based on FFMPEG, image interfacebased on OpenCV and GUI designed based on Qt SDK in Linux.In the image processing module, an efficient algorithm using inter frameinformation is designed and realized in this article. This algorithm gets theforeground image by the background subtraction based on threshold. Afterthe fixing module, the elevator waiting people targets can be extracted byconnected component detection. After that, we divide the cluster result into2parts–single person and group people. We use the ‘anti-split’ algorithm andmultiple human segmentation algorithm based on the overlap area of oneobject between two frames, to recognize and track the targets. Finally, we canget the number of waiting people. Experiments conducted in different place,the accuracy kept above80%, and the speed kept between15frames persecond and20frames per second with the stable illumination. Theexperiment result shows that the system can meet the demands of elevator-control-system.Also, this article has analyzed the shortcomings of the algorithmmentioned above. Because background subtraction needs a stableillumination, there will be some errors in algorithm designed using interframe information in some particular cases. To make the system more robust,an assist algorithm based on Mean Shift segmentation and color-histogramhead tracking is presented in this article. Through the experiment result, itsaccuracy has been proved and can meet the demand of our system.