红外技术, 2018, 40 (12): 1182, 网络出版: 2019-01-23
多步预测融合 Mean-Shift的运动目标跟踪算法研究
Study on Motion Target Tracking Algorithm Based on Mean-Shift and Multi-step Prediction
Mean-Shift算法 多步预测 运动目标跟踪 Mean-Shift algorithm Bhattacharyya coefficient Bhattacharyya coefficient multi-step prediction motion target tracking
摘要
对运动目标跟踪时,主流 Mean-Shift(均值偏移)算法对环境的影响较为敏感。针对目标遮挡时准确跟踪这一问题,提出了多步预测融合 Mean-Shift的优化运动目标跟踪算法。在目标跟踪的过程当中采取 Bhattacharyya coefficient(巴氏系数)辨别目标是否出现了遮挡。当目标产生遮挡的情况,采取多步预测算法,根据目标前一帧的特征信息对下一帧中目标位置信息进行判断。当运动目标离开遮挡时,则继续采取 Mean-Shift实施后续跟踪。通过对不同场景下的视频序列实行测试,其结果表明该算法可以对发生遮挡后的目标进行连续、稳健的跟踪。
Abstract
The mainstream Mean-Shift(mean shift) algorithm is more sensitive to environmental impacts when tracking moving targets. Aiming at the resolution of the problem of accurately tracking target occlusion, an optimal moving target tracking algorithm based on multi-step prediction fusion mean shift is proposed. In the process of target tracking, the Bhattacharyya coefficient is used to discern whether the target has occlusion. In the case of target occlusion, a multi-step prediction algorithm is adopted to determine the target position information in the next frame according to the feature information of the previous frame of the target. When the target leaves the occlusion, the algorithm continues to follow Mean-Shift for subsequent tracking. The video sequences in different environments are tested, and the results show that the algorithm can continuously and robustly track the target after occlusion.
于晓明, 李思颖. 多步预测融合 Mean-Shift的运动目标跟踪算法研究[J]. 红外技术, 2018, 40(12): 1182. YU Xiaoming, LI Siying. Study on Motion Target Tracking Algorithm Based on Mean-Shift and Multi-step Prediction[J]. Infrared Technology, 2018, 40(12): 1182.