基于孪生区域建议网络的无人机目标跟踪算法

Target Tracking Algorithm Based on Siamese Region Proposal Network for UAV

在无人机跟踪过程中,遮挡、光照变化、背景干扰等影响会导致跟踪目标丢失。基于SiamRPN算法提出一种无人机目标跟踪算法。通过在网络中加入空间条带池和全局上下文模块建立远程上下文关系,以适应不同的跟踪场景。同时利用改进交并比的计算方法提取目标特征,并回归精准的预测框。在UAV123数据集上的实验结果表明,相比SiamRPN、SiamFC、SAMF等算法,该算法的跟踪性能较优且具有较强的鲁棒性,尤其在背景干扰环境下,其精确率和成功率较SiamRPN算法分别提升了6.54%和11.63%。

When performing target tracking tasks,Unmanned Aerial Vehicles(UAV)often lose their targets due to occlusion,illumination changes and background interference.Based on the SiamRPN algorithm,a target tracking algorithm is proposed for UAV.The algorithm first adds a strip pooling and a global context module to the network,which establishs remote context and enables it to adapt to different tracking scenarios.Then the calculation method based on cross-comparison optimization is used to complete the feature extraction of the target,and to return an accurate prediction boundary frame.The experimental results on the UAV123 public data set benchmark show that compared to SiamRPN,SiamFC,SAMF algorithms,the algorithm exhibits better tracking performance and high robustness.Compared to SiamRPN algorithm,the accuracy rate and success rate of the algorithm are improved by 6.45%and11.63%respectively in the background interference.