基于三维视觉的动态目标轨迹还原方案设计

Design of dynamic object trajectory restoration scheme based on 3D vision

针对单个动态目标的轨迹还原问题,提出了基于视觉和卷积神经网络的移动平台解决方案。搭建了移动设备平台,设计了目标物轨迹从相机参考坐标系到世界参考坐标系的转换接口。引入高性能NCNN推理框架和视觉惯性里程计分别进行目标检测和定位,同时为了降低计算需求,设计了有序的多线程任务。对轨迹中的异常值点给出了剔除方案。实验结果表明:所提出的方案在计算能力有限的I5微型移动主机上可以以30FPS的速度追踪动态目标物,满足实时性要求并还原其三维轨迹,其精度高,具有实际工程应用价值。

Aiming at the problem of trajectory restoration of single dynamic target,a mobile platform solution based on 3D vision and conventional neural network was proposed.The mobile device platform was built.The conversion interface of object trajectory from camera reference coordinate system to world reference coordinate system was designed.High performance NCNN reasoning framework and visual inertial odometer were introduced to detect and locate targets respectively.In order to further reduce the computing requirements,an orderly multithreading task was designed.The outlines in the trajectory were eliminated with our method.The experimental results show that this scheme can track the dynamic object at the speed of 30 FPS on the I5 micro mobile host with limited computing power,meet the real-time requirements,and restore its three-dimensional trajectory.The accuracy has practical engineering application value.