基于核相关滤波的深空目标跟踪算法

Space Target Tracking Algorithm Based on Kernelized Correlation Filter

随着天基红外探测器接近目标,目标在焦平面的成像会由小变大,在近距离时以面目标形式成像,随着距离的接近,目标的尺度变化会越来越剧烈。针对经典核相关滤波算法无法应对尺度变化及遮挡的问题,论文提出了一种融合尺度滤波的核相关滤波跟踪算法。在位置滤波器预测目标位置之后,由尺度滤波器计算尺度响应以解决目标尺度变化问题。针对目标跟踪过程中的遮挡问题,提出了一种遮挡判别依据,遮挡情况下停止更新模板以实现稳定、鲁棒的跟踪。通过对比实验表明,该算法可以有效地应对尺度变化及遮挡情况,且帧率可以达到120 FPS,满足实时要求。

As the infrared detector approaches the target,the imaging of the target in the focal plane will change from small to large.At close range,it will be imaged in the form of area target.With the approach of distance,the scale change of the target will become more and more intense.Aiming at the problem that the classical kernelized correlation filter algorithm cannot cope with the scale-variable and occlusion,this paper proposes a kernelized correlation filters tracking algorithm that integrates scale filtering.Af-ter the position filter predicts the target position,the scale response is calculated by the scale filter to solve the target scale change problem.Aiming at the occlusion problem in the process of target tracking,an occlusion discrimination basis is proposed,which stops updating the template during occlusion to achieve stable and robust tracking.Comparative experiments show that the algorithm can effectively deal with scale-variable and occlusion,and the single frame processing speed reaches 120 FPS,which meets the re-al-time requirements.