引用本文:刘洪彬,常发亮,刘春生,等.基于时空渐进特征模型的抗遮挡多目标跟踪[J].控制与决策,2019,34(10):2171-2177
【打印本页】   【HTML】   【下载PDF全文】   查看/发表评论  【EndNote】   【RefMan】   【BibTex】 附件
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被:浏览 42次   下载 50 本文二维码信息
码上扫一扫!
分享到: 微信 更多
基于时空渐进特征模型的抗遮挡多目标跟踪
刘洪彬,常发亮,刘春生,梁付新
(山东大学控制科学与工程学院,济南250061)
摘要:
针对多目标跟踪过程中存在目标相互遮挡的问题,提出一种基于时空渐进特征模型的抗遮挡多目标跟踪方法.该方法根据目标的关联状态、重叠情况、深度顺序构建遮挡推理模型,在线识别被遮挡目标, 并对其进行基于时空渐进特征模型的目标再检测.首先,利用目标特征模型中的空间位置参数确定搜索区域;然后,计算目标模型与检测响应之间的表观特征相似度,引入一种自适应阈值,并选择相近程度较高的检测响应为候选量测;最后,采用时间特征匹配的方式获得目标真实检测响应,完成多目标跟踪的任务.通过MOT15和MOT17数据集中的实验结果表明,所提出方法能够有效地识别被遮挡目标,可在多种复杂场景下实现相互遮挡目标的鲁棒性跟踪,相对于现有方法具有较高的准确性、精度和轨迹完整性.
关键词:  遮挡推理  时空渐进  目标模型  特征匹配  目标再检测  抗遮挡  多目标跟踪
DOI:10.13195/j.kzyjc.2018.0156
分类号:TP391
基金项目:国家自然科学基金项目(61673244, 61703240).
Anti-occlusion multi-target tracking with progressive spatio-temporal feature model
LIU Hong-bin,CHANG Fa-liang,LIU Chun-sheng,LIANG Fu-xin
(School of Control Science and Engineering,Shandong University,Ji'nan250061,China)
Abstract:
Multi-target tracking in inter-occlusion scene is a challenging issue. The paper proposes an anti-occlusion multi-target tracking method based on the progressive spatio-temporal feature model. The method establishes an occlusion reasoning model according to the associated states, overlap and depth of the target. The occluded target is identified by the occlusion reasoning model and re-detected by the progressive spatio-temporal feature model. Firstly, search region is determined by the spatial position parameter in the target feature model. Then, the similarity between the target model and detection is calculated, an adaptive threshold is presented, and the detection with a similarity higher than the threshold is considered as candidate measurement. Finally, the real detection of the target is obtained by matching the temporal feature, and multi-target tracking is completed. Experimental results conducted on MOT15 and MOT17 benchmarks show that the proposed method can effectively identify the occluded target and be robust for tracking inter-occlusion targets in several complicated scenes. Compared with the state-of-the-art methods, the proposed method achieves outstanding performance in accuracy, precision and trajectory completeness.
Key words:  occlusion reasoning  progressive spatio-temporal  feature matching  target re-detection  anti-occlusion  multi-target tracking

用微信扫一扫

用微信扫一扫