动态多帧视频序列图像局部特征智能滤波仿真

Simulation of Intelligent Filtering of Local Features in Dynamic Multi-frame Video Sequence Images

传统目标跟踪算法易受光照不均的影响,导致跟踪成功率低且时效性差。为解决多帧视频序列图像的目标跟踪问题,基于MWF中值加权滤波,提出一种MWF-KCF目标跟踪算法。算法首先采用高斯核的MWF算法,基于矩形与圆形双模板对视频图像进行优化处理,增强局部特征提取时的辨识度;然后分别提取并融合梯度HOG特征与空间TEF特征,提高目标的可识别性;接着采用KCF算法对图像目标进行回归框构建,并基于APCE回归更新准则不断优化目标框的位置;最后基于UFD数据集,完成多帧视频序列图像的MWF-KCF目标模型构建。在特征滤波的仿真结果显示出,MWF滤波算法能够显著提高目标跟踪的成功率,同时具有不逊色其它滤波算法的时效性;对比仿真结果显示出,与4.3节中的四类基线算法模型相比,MWF-KCF模型在UFD数据的测试集中的跟踪准确率最高,平均提升了10.66%。综上,MWF-KCF目标跟踪算法通过滤波解决光照影响,增加特征辨识度,提升目标跟踪的成功率,在智能监控领域具有一定的仿真价值。

Traditional target tracking algorithms are vulnerable to uneven illumination,resulting in low success rate and poor timeliness.In order to solve the problem of target tracking in multi-frame video sequences,MWF-KCF target tracking algorithm is proposed based on the MWF median weighted filtering algorithm.Firstly,the algorithm uses the Gaussian kernel MWF algorithm to optimize the video image based on rectangular and circular double templates to enhance the recognition of local feature extraction,and then extracts and fuses the gradient HOG feature and spatial TEF feature respectively to improve the recognizability of the target;Then the KCF algorithm is used to construct the regression box of the image target,and the position of the target box is continuously optimized based on the APCE regression update criterion.Finally,the MWF-KCF target model of multi-frame video sequence images is constructed based on the UFD data set.The simulation results of feature filtering experiments show that the MWF filtering algorithm can significantly improve the success rate of target tracking,and has the same timeliness as other filtering algorithms;The simulation results of comparative experiments show that,compared with the four baseline algorithm models in Section 4.3,the MWF-KCF model has the highest tracking accuracy in the test set of UFD data,with an average increase of 10.66%.To sum up,the MWF-KCF target tracking algorithm solves the influence of illumination through filtering,increases the recognition of features,and improves the success rate of target tracking,which has certain simulation value in the field of intelligent monitoring.