项目名称: 基于稀疏表达和主动漂移纠正的视觉目标跟踪算法研究
项目编号: No.61203270
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 自动化学科
项目作者: 范保杰
作者单位: 南京邮电大学
项目金额: 26万元
中文摘要: 视觉目标跟踪对智能视频监控,机器人导航等具有重要意义。由于目标外观和运动模式的不确定性,以及周围环境的变化,给运动目标跟踪的研究带来了极大的挑战。此项目申请拟首次应用稀疏表达理论和主动漂移纠正思想建立面向复杂场景的鲁棒准确快速的目标跟踪算法。表达上,我们通过民主融合策略融合目标的时空域信息,并考虑其拓扑结构信息;模型上,我们首次提出具有判别能力的在线字典选择模型,在保证具有较小重建误差的基础上,突出字典的判别能力。在此基础上,我们将主动漂移纠正信息引入基于稀疏表达的目标跟踪过程中,并对跟踪中的误差进行监控,及时地对目标漂移进行纠正。并结合在线字典更新机制,选择合适的更新时机,有效减少跟踪过程中的误差累计,实现在复杂场景下对目标持续稳定准确的跟踪。我们的研究思路具有原创性,并且初步的研究成果证明了其可行性。本项目的研究成果对视觉目标跟踪具有重要的理论意义和应用价值。
中文关键词: 鲁棒目标跟踪;在线学习及更新字典和分类器;主动漂移纠正;总的跟踪算法架构;归一化联合度量
英文摘要: Visual target tracking is significant for intelligent video surveillance, robot navigation. Due to the uncertain of the appearance and movement pattern of target as well as the surrounding environment change, this will bring great challenges for the research of moving target tracking. This project is the first application of sparse representation and active drift correction to establish the robust, quick, accurate target tracking algorithm in complex scenes. In expression, we integrate the time-space information of the target through democratic integration strategy, and consider its topology information. For model, we first propose the online dictionary selection model, which has the discriminative ability. On the basis of ensuring the small reconstruction error, highlight the discriminative power of the dictionary. On this basis, we introduce the drift correction information into the process of target tracking based on sparse representation, and monitor the tracking error, correct the target drift in time. Beside, we combine the online dictionary update mechanism to select appropriate update chance, reduce the accumulated error in the tracking process effectively, and achieve stable and accurate tracking. Our idea is original, and the preliminary research proves its feasibility.
英文关键词: robust object tracking;online learn and update dictionary and classifier;active drift correction;total tracking algorithm framework;normalized colliborated metric