自适应融合多层卷积特征的目标跟踪算法  

Adaptive fusion multi-layer convolution features algorithm for object tracking

作　　者：尚振宏[1] 王娜[1] 刘辉[1] SHANG Zhenhong;WANG Na;LIU Hui(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,云南昆明650500

出　　处：《传感器与微系统》2020年第6期135-138,142,共5页Transducer and Microsystem Technologies

基　　金：国家自然科学基金资助项目(61462052)。

摘　　要：针对早期的卷积神经网络使用空间分辨率较低的全卷积层输出作为特征提取层,无法精确定位目标的问题,对VGG-19网络中的多个卷积层进行分析。提出使用第一和第五卷积层作为特征提取层。通过相关滤波器学习,得到不同的模板;并且对所得到的响应图(response map)依据APCE测量方法计算权重,自适应加权融合响应图来确定目标的最终位置;再通过提取目标多个尺度的方向梯度直方图(HOG)特征估计目标的最佳尺度。实验选取公开测试视频集中的20段视频序列与其他跟踪算法进行对比。Aiming at the problem that early convolutional neural networks using the output of full convolution layer which have lower spatial resolution as the layer of feature extraction,that can’t locate target accurately,taking the output of the first layer and the fifth layer as the layer of feature extraction is proposed. Many convolutional layers in VGG-19 network are analyzed. Templates can get,through learning of the correlated filters and the weight of response map can be calculated according to the average peak to correlation energy( APCE).After that,locate the final position of the target,by adaptively weighting and fusing response map. Histogram of oriented gradient( HOG) features are used to estimate the target scale. The experiment chooses 20 video sequences in the set of public test video and compared with other object tracking algorithms.

关 键 词：卷积特征 相关滤波 自适应融合 尺度估计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]