增强现实中基于自然特征的实时跟踪方法

Natural feature-based real time tracking method for augmented reality

在增强现实系统的复杂场景中,对目标的实时跟踪受到场景中诸多因素的制约,导致实时跟踪方法效率低且不准确,为此提出一种基于自然特征的实时跟踪方法。设计了一种螺旋分割模型,对捕获的图像进行螺旋分割,利用SURF算法在分割子块中提取特征点,并进行匹配。在对目标进行跟踪定位时,利用前一帧来预测当前帧目标出现的位置,以减少SURF算法的扫描区域,加速系统运算效率。实验中分别对场景光线强弱、视点和仿射变化以及目标被部分遮挡等不同情况进行测试,该方法均表现出较高的跟踪效率。

In complex scenes of augmented reality systems ,the real-time tracking of a target is made difficult by many factors within the scene and as a result many current real time tracking methods are low in efficiency and accuracy .To solve the prob-lem ,a natural feature-based tracking method was proposed that utilized a spiral model for segmenting captured images .Once an image was segmented ,feature points in the segmented blocks were extracted using the SURF algorithm and then matched with those in the template image set .Whilst tracking the target ,the position in the current frame was predicted according to the posi-tion in the previous frame in order to reduce search areas and improve the algorithm efficiency .Under situations of varying light intensity and viewpoints ,affine transformation and partially occluding targets ,experimental results show this method has high tracking efficiency .