摘要
视频目标跟踪是计算机视觉领域一个重要的研究方向,在公共交通、无人机、军事目标定位等诸多领域有着很重要的实际应用价值.传统的跟踪器算法在应对现实中的复杂场景具有很大瓶颈,伴随着大数据时代的到来,深度学习技术凭借着强大的特征自学习能力在图像分类、目标检测等计算机视觉领域掀起了研究热潮,同时也为目标跟踪领域的研究提供新的思路.基于各种深度神经网络模型的跟踪算法已经开始应用在目标跟踪问题中,并且在性能上取得了良好的效果.本文首先简要回顾了的传统目标跟踪算法的相关工作流程,其次,重点阐述了深度学习技术在目标跟踪领域中的应用特点,并同时对算法进行分类讨论.最后,总结了深度学习在目标跟踪领域的技术难点与未来的发展趋势.
Video object tracking is the significant research fields of computer vision,with wide application and real value in many ways such as public transport、unmanned aerial vehicle （UAV） and military target location.In a real and complex scenario,traditional object tracking algorithms are challenged by a variety of difficulties.With the advent of the era of big data,deep learning has strong characteristics of adaptive learning ability so that became a vigorous a research campaign in the fields of image classification、object detection,meanwhile,provide the enlightenment for the object tracking.All kinds of tracking algorithms based on deep learning has been implemented in object tracking task,and achieved good performance.This paper first reviews the workflow of the traditional object tracking algorithm.Secondly,we expounds the principal distinguishing characteristics of the deep learning technology applicate in tracking field and classifying the recently algorithm based on deep learning.Finally,we discuss the technical difficulties and future development trend.
出处
《小型微型计算机系统》
CSCD
北大核心
2018年第2期315-323,共9页
Journal of Chinese Computer Systems
基金
华侨大学研究生科研创新能力培育计划项目(1511314020)资助
华侨大学科技创新能力提升计划"中青年教师科技创新资助计划"项目(ZQN-PY210)资助
国家自然科学基金面上项目(61572205
61673185
61370006
61673186)资助
福建省自然科学基金项目(2015J01257)资助
关键词
目标跟踪
深度学习
神经网络
视频处理
object tracking deep learning neural network video processing
