光学学报, 2019, 39 (7): 0715002, 网络出版: 2019-07-16   

基于多层深度卷积特征的抗遮挡实时跟踪算法 下载: 1172次

Real-Time and Anti-Occlusion Visual Tracking Algorithm Based on Multi-Layer Deep Convolutional Features
作者单位
1 中国科学院国家空间科学中心复杂航天系统电子信息技术重点实验室, 北京 100190
2 中国科学院大学, 北京 100049
摘要
为提高复杂场景中目标跟踪算法的准确性与实时性,提出一种基于多层深度卷积特征的抗遮挡实时目标跟踪算法。针对目标跟踪任务,先对深度卷积网络VGG-Net-19进行微调,再提取目标区域的多层深度卷积特征。根据相关滤波框架构建位置相关滤波器,确定目标中心位置。设计尺度相关滤波器对目标区域进行不同尺度采样,确定目标尺度。目标遮挡时,采用阶段性评估策略进行模型更新与恢复,解决模型误差积累问题。选取目标跟踪评估数据集OTB-2015(100组视频序列)与UAV123(123组视频序列)进行测试。实验结果表明,本文算法具有更高的准确性,能够适应目标遮挡、外观变化及背景干扰等复杂情况,平均速度为29.6 frame/s,满足目标跟踪任务的实时性要求。
Abstract
In order to improve the accuracy and real-time performance of visual tracking in complex scenes, a real-time and anti-occlusion visual tracking algorithm based on multi-layer deep convolutional features is proposed. For the visual tracking task, the deep convolutional networks VGG-Net-19 are fine-tuned, and then the multi-layer deep convolutional features of the target region are extracted from the adjusted model. The location correlation filters are constructed to determine the target center position. In order to determine the target scale, a scale correlation filter is performed to sample multi-scale images surrounding the target region. When the target is occluded, the stage evaluation strategy is used to update and recover the model, which solves the problem of template error accumulation. The experimental results on the tracking benchmark OTB-2015 which concludes 100 video sequences and UAV123 which concludes 123 video sequences show that the proposed algorithm has higher accuracy and can adapt to complex situations such as target occlusion, appearance change and background clutters. The average speed is 29.6 frame/s, which meets the real-time requirements of the visual tracking task.

崔洲涓, 安军社, 崔天舒. 基于多层深度卷积特征的抗遮挡实时跟踪算法[J]. 光学学报, 2019, 39(7): 0715002. Zhoujuan Cui, Junshe An, Tianshu Cui. Real-Time and Anti-Occlusion Visual Tracking Algorithm Based on Multi-Layer Deep Convolutional Features[J]. Acta Optica Sinica, 2019, 39(7): 0715002.

本文已被 2 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!