基于ELM和核相关滤波器的自适应目标跟踪算法

Adaptive target tracking algorithm based on ELM and KCF

针对传统的核相关滤波器(KCF)跟踪算法在目标快速运动、尺度变化和遮挡情况下通常会导致跟踪失败的问题,在传统的KCF算法的基础上引入极限学习机(ELM),提出一种基于ELM和KCF的自适应目标跟踪方法。根据过去时刻的目标位置信息,利用ELM预测出当前帧目标的可能位置;在该位置上以目标区域为基础进行多尺度目标图像特征采样,通过KCF确定目标的最终位置和最佳尺度;通过计算目标位置响应图的振荡程度来自适应地改变模型的更新速率。在36组公开视频序列上对所提算法与6种当前主流的相关滤波跟踪算法进行了实验,所提算法取得了最好的跟踪精度和成功率,能够有效处理目标遮挡、快速运动和尺度变化等问题,具有较为重要的理论研究和应用价值。

Aiming at the problem that traditional kernel correlation filter(KCF)algorithm often fails in object fast movement,scale changes and occlusion case,an adaptive target tracking method based on extreme learning machine(ELM)and KCF is proposed.Firstly,according to the target location information of the past time,ELM is used to predict the possible position of the target in current frame.Then,the feature sampling of multi-scale target image is carried out based on the target area at the predicted position,and the final position and optimal scale of the target are determined by KCF.Finally,the updating rate of the model is adaptively changed by calculating the oscillation degree of response map at the target location.The proposed algorithm and 6 state-of-the-art algorithms based on correlation filter are tested on 36 public video sequences,the proposed algorithm achieves the best tracking precision and success rate,and it can effectively handle object occlusions,rapid movement and scale changes.Furthermore,it has important theory research and application value.