关键词: 平板图案跟踪, 增强现实, 角点匹配算法, 非线性优化

Abstract:

An approach for planar pattern tracking was introduced. The tracker could locate a planar pattern accurately in real time using a Webcamera. The tracking result could be used for computing camera's position and placing virtual object in the video. Some common tracking failures caused by "motion blur", "fast motion", "small target" were discussed. An algorithm, combined "LK tracking" techniques with "Hungarian algorithm-based feature correspondence", was used to solve such problems. A method based on non-linear optimization is used to suppress the influence of noise, when computing the camera's position. The experiments show that this method can track the planar pattern robustly in some difficult situations. The curve of the camera's computed position is smooth and accurate.

Key words: planar pattern body tracking, augmented reality, corner correspondence, nonlinear optimization