摘要
目标跟踪是计算机视觉领域研究的热门话题,在实践中有着广泛的应用。主要针对三维单目标跟踪问题进行研究,考虑到三维目标检测已有较好表现,将通过视锥点云网络得到的三维目标检测结果与TLD目标跟踪算法相结合,并改进TLD算法的检测模块,构造了基于TLD的三维单目标跟踪算法,输入RGB-D图像序列,输出目标的三维跟踪结果。在KITTI数据集中测试结果表明该算法在三维目标跟踪上的可行性。
Target tracking is a hot topic in the field of computer vision and has a wide range of applications in practice.The 3D single-target tracking problem is mainly studied in this paper.Considering that the 3D target detection has good performance the 3D target detection result obtained by using the frustum point cloud network is combined with the TLD target tracking algorithm and the detection module of the TLD algorithm is improved.A 3D single-target tracking algorithm based on the TLD algorithm is proposed.The input is the RGB-D image sequence and the output is the 3D tracking results of the target.The test results in the KITTI data set prove the feasibility of the algorithm in 3D target tracking.
作者
赵斌
王江卓
李建勋
ZHAO Bin;WANG Jiangzhuo;LI Jianxun(Project Management Center of Air Force Equipment,Beijing 100843,China;Department of Automation Shanghai Jiaotong University,Shanghai 200240,China)
出处
《电光与控制》
CSCD
北大核心
2020年第2期22-28,共7页
Electronics Optics & Control
基金
国家自然科学基金(61673265)
