摘要
视觉目标跟踪是计算机视觉领域的重要研究方向之一,在战场侦察、视频监控、自动驾驶和视频分析等军事和民用领域得到广泛应用.近年来目标跟踪算法已经取得一系列进展,但由于在实际跟踪过程中存在着复杂的目标和背景变化,稳定的目标跟踪仍面临很多挑战.本文首先介绍了实际跟踪场景中出现的困难挑战;其次,主要针对背景相似干扰、旋转变化、遮挡、尺度变化等难点问题分别就特征提取、观测模型和模型更新三个方面对具有代表性的判别式目标跟踪方法进行深入阐述和分析;随后,在OTB2015数据集上对25种典型跟踪算法进行了实验对比与分析;最后,对视觉目标跟踪技术的发展方向进行展望.
Abstract
Visual object tracking is a fundamental problem in computer vision.It has been widely used in civilian and military fields,such as battlefield reconnaissance,video surveillance,automatic driving,video analysis,and many other areas.In recent years,although the object tracking algorithm has made great progress,stable object tracking is still a challenging task due to random target changes and complex scenarios.Firstly,the difficulties and challenges in actual tracking scenarios are introduced in this paper.Then,aiming at the background clutter,rotation changes,occlu-sion,and scale changes,the representative discriminative object tracking algorithms are summarized and analyzed from the perspective of feature extraction,observation model,and model update mechanism.Subsequently,25 typical track-ing algorithms are evaluated and analyzed on OTB2015 database.Finally,the further research directions are prospec-ted.
基金项目
基础前沿科技创新项目(KYSZJXXXXX006)
维普
万方数据