基于预判式学习更新策略孪生全卷积网络的目标跟踪

TARGET TRACKING BASED ON SIAMESE FULL-CONVOLUTION NETWORK WITH PREJUDGMENT LEARNING UPDATE STRATEGY

针对孪生全卷积网络缺乏有效的模型更新策略,跟踪目标周围存在相似目标干扰容易出现跟踪丢失的问题,提出一种基于预判式学习更新策略孪生全卷积网络的目标跟踪算法。确定目标模板和搜索区域的直方图置信度估计;模拟学习率的自重启机制,由给定正确标注初始化学习模块;根据置信度估计决定预判式学习模块更新,实现跟踪目标和相似目标的有效区分。实验结果表明,该算法具有良好的跟踪效果,在满足实时性跟踪的基础上,具有很好的跟踪精度和成功率。

Aiming at the lack of effective model updating strategy for the Siamese full-convolutional network and the problem of tracking loss due to the similar target interference around the target,a target tracking algorithm based on the prejudgment learning update strategy for the Siamese full-convolutional network is proposed.The histogram confidence estimate for the target template and the search region was determined;the self-restart mechanism of the simulation learning rate was initialized by the given correct annotation;the prejudgment learning module was updated according to the confidence estimation to achieve effective differentiation between the tracking target and the similar target.The experimental results show that the algorithm has good tracking effect and good tracking accuracy and success rate on the basis of real-time tracking.