一种基于无人机云台的视觉目标跟踪方法

Visual Object Tracking System Based on UAV PTZ

鉴于无人机视觉目标跟踪时会遇到目标遮挡、目标尺度变化等问题,同时目标跟踪方法复杂度受到云台芯片算力的严格限制,提出了一种基于无人机云台的视觉目标跟踪方法。基于多个特征相关滤波器的自适应权重融合来提升目标位置预测的精度。引入长宽相互独立变化的尺度变化池策略,解决无人机目标跟踪过程中目标尺度估计问题。通过设置模板检测和轨迹预测模块来有效地对目标跟踪过程中的遮挡状态进行检测和处理。该方法可在无人机云台芯片中实现实时跟踪,且在公开无人机目标跟踪数据集和自采集数据集中取得了很好的跟踪效果,与基线方法相比,成功率提升了10.7个百分点,准确率提升了3个百分点。

Considering the issues of target occlusion and target scale changes encountered in UAV visual target tracking,and the complexity of the target tracking method that is strictly limited by the computing power of the UAV PTZ chip,a visual target tracking method based on UAV PTZ is proposed.Adaptive weight fusion based on multiple feature correlation filters is used to improve the accuracy of target position prediction.A scale change pool strategy with independent changes in length and width is introduced to solve the problem of target scale estimation in UAV target tracking.The occlusion status during the target tracking process is effectively detected and processed by setting up template detection and trajectory prediction modules.This method can achieve real-time performance in UAV PTZ chip,and achieved good performance in both public UAV target tracking datasets and self collected datasets.Compared with the baseline method,the method improves the success rate by 10.7% and improves the accuracy by 3%.