动态场景下AGV视觉目标跟踪研究

Visual Target Tracking of AGV in Dynamic Scene

在动态场景中,AGV借助相关滤波原理进行跟踪是一种非常重要的目标跟踪方法,跟踪方法必须具有良好的准确性和实时性。首先介绍对两种主流的相关滤波跟踪算法KCF和SA进行阐述,将SA算法的尺度策略加以改进并融入KCF算法中,结合颜色特征,提出一种改进尺度自适应的Sa-KCF算法。通过设计实验,在OTB-2013数据集中选取动态场景序列,对比测试三种算法的跟踪准确性和实时性。实验表明,改进的Sa-KCF算法在综合准确性和运行效率方面占有优势,能够较好地满足AGV在视觉跟踪中鲁棒性和实时性的要求。

In the dynamic scene,it is a very important tracking method to track the target with the help of the correlation filtering principle for AGV. The tracking method must have good accuracy and real-time performance. Firstly,two mainstream tracking algorithms based on correlation filtering KCF and SA are described. A scale-adaptive Sa-KCF algorithm combined with color features is proposed by improving the scaling strategy of SA algorithm and integrating it into KCF algorithm. Through the design experiments,dynamic scene sequences are selected in OTB-2013 data set to test the tracking accuracy and real-time performance of the three algorithms. Experiments show that Sa-KCF algorithm has advantages in comprehensive accuracy and operation efficiency,and it can meet the requirements of robustness and real-time in visual tracking.