追踪过程中的无人机目标重定位方法研究

Research on UAV Target Relocation Method in Tracking Process

无人机追踪过程中,遮挡、姿态变化等易引起目标丢失,而核相关滤波算法(KCF)缺乏样本提取能力无法重新定位目标,导致追踪失败。对此,提出一种利用轨迹预测和空间感知的无人机追踪方法来重定位目标。通过YOLOv3算法和双目立体视觉,实现样本提取、目标重定位和优化尺度判定机制,并引入扩展卡尔曼滤波器,预测目标位置,消除目标丢失时无人机失去引导的缺陷。最后,通过gazebo物理引擎平台的仿真实验和实际环境下的追踪实验,证实了算法的有效性。

Target loss,susceptible to occlusion and changeable attitude,is a common problem in UVA’s actual tracking,while it is difficult to resolve by resorting to the traditional Kernel Correlation Filtering(KCF)algorithm which lacks the capacity of sample extraction.Considering this,an improved target tracking algorithm was proposed featuring with trajectory prediction and spatial sensing.YOLOv3 algorithm and Binocular Stereo Vision were employed to identify the target and calculate its spatial position.This step not only contributes to the sample extraction of KCF algorithm and target relocation,but also optimizes the mechanism to determine confidence level and target scale.The target position was predicted on spatial coordinates after introducing the extended Kalman filter,which remedies the absent guidance for UVA when losing target.Simulation experiment was conducted on the platform of gazebo physical engine and the algorithm was experimentally verified in the actual environment.