People Tracking and Counting
Disguise monitoring system. Counting people passing through monitored areas remains a critical research topic in this field, primarily relying on background subtraction processes. The method faces two major challenges: dynamic background model estimation and shadow removal. To address these, a bidirectional people counting algorithm is proposed. For developing robust counting systems, Gaussian Mixture Models (GMM) are employed to characterize background scenes. However, this algorithm lacks classification capability for detecting shadows in moving foreground objects. Performance enhancement is achieved by integrating color models with background models, improving motion object detection through shadow elimination from foreground elements. A multi-class feature-based tracking algorithm handles occlusion issues in multi-object tracking, while bidirectional counting improvement requires multi-level backward tracking procedure development.