Robust Visual Detection and Recognition of Moving Objects in Real-Time

摘要

This report covers research in three areas of vision technology: the detection,tracking and recognition of objects. The focus of the Phase I research effort was to demonstrate the feasibility of the technology for security and teleconferencing applications. To this end, a system capable of detecting, tracking and identifying human subjects was developed. For object detection and tracking, differential motion segmentation is employed, and both region-based and boundary-based algorithms were developed for this purpose. The recognition module is based on an approach to matching flexible, poorly modeled objects that has the potential speed of traditional indexing methods. The design is an original combination of moment-based image representation, relational match analysis, efficient indexing and robust transformation estimation. The feasibility of all the designs are demonstrated experimentally in the context of the applications.