基于粒子滤波与自然特征匹配的虚实配准方法

A Registration Method Based on Particle Filter and Natural Feature Matching

提出了一种基于粒子滤波与多尺度自然特征点匹配相结合的虚实配准方法.该方法具有如下特点:(1)无需向场景中添加任何人为标识即可正常工作;(2)采用系统自适应传递模型以保证系统在摄像机抖动或运动路径突然变化等情况下能有效完成虚实配准;(3)采用简化尺度不变特征变换和尺度预测技术完成特征点匹配.实验结果表明,该方法能适应摄像机抖动、场景部分遮挡、光照以及视点和视角大幅度变化等情况,并适应室内和户外环境,且具有较高的配准精度.

A Registration Method based on particle filter and natural feature matching was proposed. The method has following advantages： （1） It can work with arbitrary geometric shapes including planar, near planar and non planar structures which really enhance the usability of AR systems. （2） The adaptive system transfer model was adopted for camera shaking and erratic motion. （3） A simplified scale-invariant features transform and scale prediction technology for completion of feature matching was introduced, which guarantee robustness and stability of the system. Experiment results show that the method can still wok well even under the circumstances of camera shaking, ocdusion and large changes in illumination, viewpoint and view angle during the entire process.