基于DPM和KCF的十字靶标检测与跟踪

摘要

An automatic detection and tracking model of cross target is established in order to realize automatic detection and tracking of cross target in this paper.There is a large amount of computation and poor real-time performance in the traditional target detection algorithm.And the target tracking requires manual calibration of the initial frame of video.To solve the problems above, a cross target detection and tracking algorithm is proposed based on Deformable Part Model (DPM) and Kernelized Correlation Filter (KCF) .First, the Histogram of Oriented Gradient (HOG) feature of cross sample set is extracted.The feature set is trained by using Latent SVM classifier, and the DPM model of target object class is generated.Then the image to be detected is traversed by sliding window search matching method.Finally, using the detected result as the starting tracking frame of the tracking algorithm, the KCF algorithm is applied to track the target quickly.When the tracking target is lost, the tracking is stopped temporarily.The positioning target is redetected with the DPM model, and then tracked with KCF algorithm.Experimental results indicate that the average frame rate detected by DPM model is 1 fps.The average frame rate of real-time detection and tracking is 40 fps combined with DPM and KCF algorithm.By using the algorithm proposed in this paper, it realizes automatic detection and real-time tracking of the target, and the detection speed is obviously higher than the traditional algorithm.After the target lost, it automatically locks and keeps on tracking, realizing the goal of long tracking of the cross target.%为了实现对十字靶标的自动检测与跟踪, 建立了十字靶标检测跟踪模型.针对目标检测中运算量大、实时性差、目标跟踪需要人工标定视频初始帧的问题, 提出了一种基于可变形部件模型 (DPM) 和核相关滤波器 (KCF) 的十字靶标检测跟踪算法.首先, 提取十字样本集的梯度方向直方图 (HOG) 特征, 采用Latent SVM分类器训练特征集, 生成十字靶标物体类的DPM模型.然后, 通过滑动窗口搜索匹配方法遍历待检测图片.最后, 将检测到的结果作为跟踪算法的起始跟踪帧, 应用KCF算法快速跟踪目标.当跟踪目标丢失时, 暂时停止跟踪, 利用DPM模型重新检测定位目标再进行跟踪.实验结果表明:采用DPM模型检测的平均帧率为1 fps, 结合DPM和KCF算法, 实时检测跟踪的平均帧率为40 fps.采用基于可变形部件模型 (DPM) 和核相关滤波器 (KCF) 的十字靶标检测跟踪算法, 实现了目标的自动检测与实时跟踪, 且检测速度明显高于传统算法, 并且在目标漂移或丢失后自动重新定位并继续跟踪, 完成十字靶标的长时间跟踪.