The aim of this chapter is to describe a new people counting technique based on spatio-temporal optical flow analysis. Using a single vertical overhead or oblique-mounted camera, our technique counts the number of people crossing a virtual line via an online blob detector and a linear classifier. In addition to the counting information, this technique identifies the speed of the observed blobs: fast, slow and stationary. It also identifies the nature of the blob: one person, two persons and group of persons. The suggested technique was validated by several realistic experiments. We showed the real-time performance and the high counting accuracy of this technique in indoor and outdoor realistic dataset.
KeywordsMovement Vector Template Match Flow Estimation Active Shape Model Count Estimation
Unable to display preview. Download preview PDF.