结合背景信息的自适应加权压缩跟踪算法

发布时间：2019-01-28 09:55

【摘要】：目的为了提高目标跟踪的准确度,提出结合背景信息的自适应加权压缩跟踪算法。方法针对目标边缘背景信息的干扰问题,提出对目标框架分块提取特征,根据区域分配权值,弱化特征提取过程中背景信息的干扰;利用正负样本特征概率分布的Bhattacharyya距离,自适应地选取区分度较大的特征进行分类器训练,提高分类器的鲁棒性;针对目标遮挡导致分类器分类不准确问题,提出设置目标遮挡检测机制,结合目标和局部背景信息对目标实现遮挡环境下的跟踪。结果与目前较流行的5种算法在6个具有挑战性的序列中进行比较,本文提出的算法平均跟踪率达到90%,平均每帧耗时0.088 6 s。结论本文算法在背景干扰,光线变换,目标旋转、形变、遮挡和复杂背景环境下的跟踪具有较高鲁棒性。
[Abstract]:Aim to improve the accuracy of target tracking, an adaptive weighted compression tracking algorithm combined with background information is proposed. Methods to deal with the interference of background information on the edge of the target, the feature is extracted from the target frame block, and the interference of the background information in the extraction process is weakened according to the weight value of the region distribution. By using the Bhattacharyya distance of the probability distribution of positive and negative samples, the classifier is trained adaptively to select the feature with a large degree of discrimination to improve the robustness of the classifier. Aiming at the inaccurate classification of classifier caused by target occlusion, a target occlusion detection mechanism is proposed, which combines target and local background information to track the target in occlusion environment. Results compared with five popular algorithms in six challenging sequences, the average tracking rate of the proposed algorithm is 90, and the average time per frame is 0.088 s. Conclusion the proposed algorithm is robust to background interference, ray transformation, target rotation, deformation, occlusion and complex background.
【作者单位】：江西理工大学信息工程学院;
【基金】：国家自然科学基金项目(61105042,61462035) 江西省青年科学家(井冈之星)培养对象计划基金项目(20153BCB23010)~~
【分类号】：TP391.41

【相似文献】