基于偏好学习的视频图像单目标跟踪算法研究

Study on Video Image Single Target Tracking Algorithm Based on Preference Learning

针对视频图像连续拍摄中不利的环境因素造成图像目标跟踪丢失的现象,本研究提出了一种基于SDAE模型和偏好学习模型结合的视频图像单目标跟踪算法。在SDAE模型中,采用卷积神经网络模型的处理规则,并对SDAE堆栈式去噪自编码器的内部排列结构进行了调整;构建的偏好学习模型,将目标跟踪问题转换为目标图像块中重叠部分的区域大小排序问题,完成了排序函数、样本之间的偏好关系、跟踪约束条件和支持向量机二分类器的设计。本研究算法与实验选取的四种目标跟踪算法相比的结果表明,本研究算法在跟踪目标成功率、目标跟踪精度和系统运行时间方面具有一定优势,目标跟踪成功率和目标跟踪精度均为89%左右。

For the phenomenon of image target tracking loss caused by adverse environmental factors in the continuous shooting of video images,a single target tracking algorithm for video images based on the combination of SDAE model and preference learning model was peoposed in this study.In SDAE model,the processing rules of convolutional neural network model were adopted,and the internal arrangement structure of SDAE stack de-noising self-coder was adjusted.The constructed preference learning model transforms the target tracking problem into the region size sorting problem of the overlapping part of the target image block,and completed the design of the sorting function,the preference relationship between samples,the tracking constraints and the support vector machine classifier.Compared with the four target tracking algorithms selected in the experiment,the results showed that this algorithm has certain advantages in the tracking the target success rate,the target tracking accuracy,and the system running time,and the target success rate and target tracking accuracy were about 89%.