一种融合卡尔曼滤波的改进时空上下文跟踪算法

发布时间：2018-07-30 08:38

【摘要】：针对时空上下文跟踪算法对快速运动和遭受严重遮挡目标的跟踪精度下降问题,提出一种融合卡尔曼滤波的改进时空上下文跟踪算法。首先人工标记目标所在的矩形区域,然后利用改进的时空上下文算法对目标进行稳定跟踪,在跟踪过程中,基于连续两帧图像灰度的欧氏距离判定目标跟踪状态,当判定目标遭受严重遮挡时,利用卡尔曼滤波进行预测估计。算法对噪声有一定的容忍度,通过降低噪声对跟踪过程的影响,能够得到更优的目标区域。仿真实验结果表明:本文算法适用于不同光照强度下高速、高机动目标的稳定跟踪,并且对目标的尺度变化和短时严重遮挡具有鲁棒性。算法帧平均耗时为34.07ms;帧几何中心平均误差为5.43pixel,比时空上下文算法减少70.2%;目标轮廓面积平均误差为13.08%,比时空上下文算法减少52.7%。
[Abstract]:An improved spatio-temporal context tracking algorithm based on Kalman filter is proposed to reduce the tracking accuracy of fast moving and severely occluded targets. Firstly, the rectangular region of the target is manually marked, and then the target is tracked stably by using the improved spatio-temporal context algorithm. In the tracking process, the Euclidean distance based on the gray scale of two consecutive frames is used to determine the tracking state of the target. The Kalman filter is used to estimate the target when it is severely occluded. The algorithm has a certain tolerance to noise. By reducing the influence of noise on the tracking process, a better target area can be obtained. The simulation results show that the proposed algorithm is suitable for the stable tracking of high speed and high maneuvering targets with different illumination intensity, and is robust to the scale change of the target and the short term severe occlusion. The average frame time of the algorithm is 34.07 Ms, the average error of the geometric center of the frame is 5.43 pixel, which is 70.2 less than the space-time context algorithm, and the average error of the target contour area is 13.08, which is 52.7 less than the space-time context algorithm.
【作者单位】：西北工业大学航天学院智能机器人研究中心;西北工业大学航天飞行动力学重点实验室;
【基金】：国家自然科学基金(11272256,61005062,60805034)~~
【分类号】：TP391.41

【相似文献】