基于KCF的目标遮挡再检测算法

Target occlusion detection algorithm based on KCF

为了解决传统的核相关滤波跟踪算法在目标遮挡时存在的问题,对核相关滤波目标跟踪算法的框架进行了研究,提出了一种融合感知哈希算法和Farneback光流算法的改进KCF算法,引入了一种目标遮挡后重新搜索定位目标的策略;利用感知哈希算法计算目标与追踪到的目标间相似度,判定目标是否发生遮挡。当目标发生遮挡时,以一定的取样间隔到历史模板中取出一帧视频图像,作为新的样本输入到KCF算法检测下一帧中目标位置,对目标进行跟踪。TB测试数据集上的检测结果表明,目标在发生遮挡的情况下,能够重新找到目标并进行准确的跟踪。

In order to solve the target sheltered existence question,this paper proposes a fusion algorithm which includes perceptual hash algorithm and Farneback opticalflow algorithm,and introduces a strategy of researching and locating targets after occlusion.Using the perceptual hash algorithm it calculates the similarity between the target and the tracked target,and to determine whether the target occludes.When the target is occluded,a frame of a video image is taken out from the historical template at a certain sampling interval and input into the KCF algorithm as a new sample to detect the target position in the next frame and track the target.The detection results on the TB test data set indicate that the target can be recovered and tracked accurately in case of occlusion.