Shopping malls and public transportation systems would like to keep track of shoppers and passengers volume by employing automatic techniques. We are developing a Real-time Stereo-Based Passenger Flow Estimation System for these needs. In this paper, the stereo-based approach that detects and tracks people from a stereo camera mounted above a door and pointing down is described. It contains three key points:1)obtaining left and right images synchronously, 2)stereo matching for real-time application, 3)applying 2-D image processing results and designing Depth criteria in a hierarchical manner, the efficient perceptual grouping of human head is realized, and with depth prediction, the accuracy of head tracking is improved. |