自适应特征融合的多尺度相关滤波目标跟踪算法

Multi-scale Correlation Filter Tracking Algorithm Based on Adaptive Feature Fusion

针对单一特征目标跟踪算法不能较好地适应复杂场景的变化,容易受跟踪目标的尺度变化、形变、遮挡以及背景混杂等影响导致跟踪失败的问题,提出一种自适应特征融合的相关滤波目标跟踪算法.首先根据目标的HOG和CN特征,采用上下文感知相关滤波框架得到2种特征下滤波响应值,并且进行归一化处理;然后按照响应值占比分配权重并线性加权融合,将得到融合后响应值用于确定目标位置;再引入尺度相关滤波器,用于估计目标尺度变化,增强尺度应变能力;最后通过设定的预定义阈值来判断位置和尺度滤波模型的更新,提高模型的更新质量.采用OTB Benchmark数据集进行实验,分别与基于相关滤波和基于上下文感知框架等11种目标跟踪算法进行对比,结果表明,该算法在精确度和成功率上均取得较为理想效果,其中精确度为82.5%,成功率为54.2%;而且在尺度变化、形变、快速运动、遮挡等复杂场景挑战下具有较好的鲁棒性.

Since the single feature target tracking algorithm cannot adapt to the change of complex scene well,it is easily affected by the influence of scale variation,deformation,occlusion and background mixed and so on,which leads to the failure of tracking.This paper proposes a multi-scale correlation filter tracking algorithm based on adaptive feature fusion.Firstly,according to the HOG and CN features of the target,the context-aware correlation filter framework was adopted to get the filtering response values of the two features,and the two response values were subsequently normalized.Secondly,according to the distribution weight of the response value and the linear weighted fusion,the fusion response value was calculated to determine the target location.At the same time,scale correlation filter was introduced to estimate target scale changes and enhance scale adaptability capacity.Finally,the quality of model updating was improved by using the predefined response threshold as the judgment condition of translation and scale filter model update.The OTB Benchmark data set is used to test the proposed algorithm,and the experiments are compared with the 11 target tracking algorithms based on the correlation filtering and the context aware framework.The experimental results show that the proposed algorithm achieves promising tracking results,where the distance precision rate is 82.5%and the overlap success rate is 54.2%,and has strong robustness in sophisticated scenarios such as scale variation,deformation,fast motion,occlusion,and so on.