| Summary: | 碩士 === 國立暨南國際大學 === 電機工程學系 === 103 === An existing approach for automatic image recognition and tracking is based on a cascade
architecture. In this architecture, the recognition part adopts the Adaboost cascade
classifier, whereas the tracking part adopts CamShift tracking and the Kalman estimator.
However, weaknesses exist in both the recognition part and tracking part. In order
to achieve a robust automatic real-time tracking system, this thesis builds a tracking system
based on the previous research. To overcome the existing weaknesses, the following
ideas are proposed in this thesis: the Adaboost classifier is combined with Mahalanobis
distance-based decision making to increase the recognition rate, and size estimation is
added to increase tracking performance. The proposed method overcomes the shortcoming
of traditional CamShift which requires manual target designation; it also overcomes
problems due to object color similarity. Moreover, an increased recognition rate of composite
objects is achieved.
|
|---|
