计算机工程与应用 ›› 2016, Vol. 52 ›› Issue (6): 134-138.

• 模式识别与人工智能 • 上一篇    下一篇

基于Gabor变换的快速跟踪算法

徐天阳,吴小俊   

  1. 江南大学 物联网工程学院,江苏 无锡 214122
  • 出版日期:2016-03-15 发布日期:2016-03-17

Fast tracking algorithm based on Gabor transformation

XU Tianyang, WU Xiaojun   

  1. School of IoT Engineering, Jiangnan University, Wuxi, Jiangsu 214122, China
  • Online:2016-03-15 Published:2016-03-17

摘要: 为了增强目标跟踪的速度和精度,提出了一种基于Gabor变换的快速跟踪算法。根据Gabor变换对人类视觉感受野良好的模拟能力,用多尺度多方向的Gabor滤波器对目标图像进行特征抽取,以此建立目标的表观模型,而后利用图像匹配的方法得到相邻帧目标位置的后验概率分布从而实现跟踪。其中在特征抽取级利用线性多通道模型将不同尺度和方向的Gabor特征融合起来,在输出级利用时频的卷积特性以FFT实现相邻帧目标位置后验概率的快速计算,充分考虑了跟踪的速度和精度。实验结果表明,该算法选用的Gabor特征对目标有准确的描述能力,以此建立的表观模型鲁棒性强;同时跟踪过程简单快速,在精度和速度上与其他前沿的跟踪算法相较有优越性。

关键词: 目标跟踪, Gabor滤波, 特征融合, 图像匹配

Abstract: In order to enhance the speed and accuracy of object tracking, a fast tracking algorithm based on Gabor transformation is proposed. According to the good simulation capability of Gabor transformation to human visual receptive field, the proposed algorithm extracts features via multi-scale and multi-orientation Gabor filters, and then realizes tracking by utilizing image matching between target model and candidates. At the feature extracting stage, a multi-channel model is used to fuse Gabor features. And at the output stage, convolution property in spatial-frequency domain is exploited to realize fast posterior distribution computation. Experimental results indicate that the proposed algorithm has good properties in accuracy and speed, and outperforms state-of-the-art methods.

Key words: visual tracking, Gabor filter, feature fusion, image matching