基于核相关滤波器的高可信度自适应融合目标跟踪算法研究

目前基于核相关滤波器的几种主流算法仍然存在目标漂移甚至跟丢的情况，本文算法在Staple: Complementary Learners for Real-Time Tracking（以下简称Staple）和Large Margin Object Tracking with Circulant Feature Maps（以下简称LMCF）两种算法基础上进行融合并改进。首先利用贝叶斯公式求解出当前帧的背景和前景的直方图，对前景直方图进行均衡化，以消除噪声；其次，利用巴氏系数计算出当前帧的背景和前景的直方图的相似度值，通过数据分析设定阈值，以此消减目标与背景过于相似的问题；再次，根据所设阈值将直方图的bin值进行自适应，以此提高跟踪的分辨率；然后利用LMCF算法求出当前帧的最大峰值和平均峰值相关能量（APCE），并根据数据分析设定合适的阈值；最后将阈值范围内的最大值和APCE值与融合系数相结合，使算法达到自适应融合。本文算法（LMCF-Staple）在公开数据集平台OTB50和OTB100上进行测试，结果显示LMCF-Staple算法的跟踪稳定性大大提高，目标的漂移情况明显减少，并且其跟踪的精确度、成功率均优于目前主流的几种算法。 

Research on high-confidence adaptive fusion target tracking algorithm based on kernel correlation filter

At present, several mainstream algorithms based on nuclear correlation filters still have target drift or even loss. The algorithm in this article is in fusion and improvement based on the two major algorithms，Staple: Complementary Learners for Real-Time Tracking (hereinafter referred to as Staple) and Large Margin Object Tracking with Circulant Feature Maps ( hereinafter referred to as LMCF). First, the Bayesian formula is used to solve the histogram of the background and foreground of the current frame, and the foreground histogram is equalized to eliminate certain noise; second, the Bhattacharyya Coefficient is used to calculate the histogram of the background and foreground of the current frame for similarity value, the threshold is set through data analysis to reduce the problem that the target and the background are too similar; again, the bin value of the histogram is adapted according to the set threshold to improve the resolution of tracking; then the LMCF algorithm is used to calculate the maximum peak value and average peak correlation energy (APCE) of the current frame, and set an appropriate threshold based on data analysis; finally, the maximum value within the threshold range and APCE value are combined with the fusion coefficient to make the algorithm achieve adaptive fusion. The algorithm in this paper (LMCF-Staple) is tested on the public data set platforms OTB50 and OTB100, the public data set platform. The results show that the tracking stability of the LMCF-Staple algorithm is greatly improved, the target drift is significantly reduced, and its tracking accuracy and success rate are better than the current mainstream algorithms.