Abstract:

The problem of object scale and direction change has always been a difficulty in the process of visual object tracking. How to deal with the change of object scale and direction effectively is an important factor to ensure the robustness of object tracking algorithm. This paper introduces the development of video object tracking, and classifies the existing object scale and direction tracking algorithms：incremental search, mean shift iteration, corner matching, region second-order moment, particle filter, correlation filter and depth learning tracking algorithm. The basic idea of various algorithms and their processing methods of scale and direction are described, focusing on the analysis of the method of using depth learning technology to deal with the change of object scale and direction, analyzing the advantages and disadvantages of various algorithms, and pointing out their applicable occasions. The future development trend of object scale and direction tracking is prospected, and the main challenges and problems are put forward, which can be used for reference for related researchers.

Key words: object tracking, scale and direction, adaptation, visual

摘要：

视觉目标跟踪过程中出现的目标尺度和方向变化问题一直是目标跟踪中的难点，如何有效处理目标尺度方向变化是保证目标跟踪算法鲁棒性的一项重要因素。介绍了视频目标跟踪发展状况，并对现有的目标尺度和方向跟踪算法进行了分类：增量式搜索、Meanshift迭代、角点匹配、区域二阶矩、粒子滤波、相关滤波器和深度学习跟踪算法。阐述了各种算法的基本思想及其尺度和方向处理方法，重点分析了利用深度学习技术处理目标尺度和方向变化的策略，分析了各种算法的优缺点，并指出了它们的适用场合。对目标尺度和方向跟踪未来发展趋势进行了展望，提出了主要挑战和难题，对相关人员的研究工作起到参考和借鉴作用。

关键词: 目标跟踪, 尺度方向, 自适应, 视觉