摘要
提出一种基于深度学习的目标跟踪算法,该算法利用深度学习中深度置信网络(Deep Belief Networks,DBN)的方法,根据输入图像标记信息,深度分析图像结构,训练得到一个分类器对各帧图像中的像素进行分类,进而得到一个二值图像作为目标分布图,然后利用Camshift算法计算出目标质心位置,实现对目标的跟踪。同时,在跟踪过程中,根据当前帧目标的位置信息,加入Kalman滤波预测,以提高目标跟踪的效率。在多个视频中对所提出的算法进行验证,实验结果表明,算法可以很好地应对光照变化、目标旋转、遮挡等多种复杂环境,实现对目标的稳定跟踪。
An object tracking algorithm based on deep learning was proposed,it uses the Deep Belief Networks（DBN）method,according to labeling information in the input image and depth analysis of the image structure,training a classifier to classify the pixels in each image frame in order to get a binary image as the target distribution map,then it uses Camshift algorithm to calculate the centroid position of the target. At the same time,in the process of tracking,it improves the efficiency of target tracking by adding Kalman filter prediction according to the position information of the object in current frame. The experimental results in multiple videos show that the algorithm can well deal with illumination changes,target rotation,occlusion and other complex environment,and achieves tracking object robustly.
作者
李克靖
孙凤梅
LI Ke-jing;SUN Feng-mei(China Electronic Technlogy Group Corporation No.58 Research Institute, Wuxi 214035, China)
出处
《电子设计工程》
2018年第11期127-131,共5页
Electronic Design Engineering
