基于KALMAN滤波和深度学习的机器人飞拍方法研究

Research on flying-shooting method of robot based on KALMAN filtering and deep learning

针对传统机器人飞拍系统存在视-控分离,导致系统实时性和效率不高的问题,提出视-控一体的设计方案,并对机器人飞拍的相关算法展开研究。为提高系统的实时性,操作系统采用基于Windows的硬实时系统,机器人控制模块和视觉模块在同一操作系统里通过共享内存的方式实现数据交互;基于改进型卡尔曼滤波算法,提出一种无传感器的精准时间触发算法,减小相机触发时机器人在同一位置的重复度误差;结合深度学习算法,提高图像处理的处理速度,保证在机器人下一节拍前完成纠偏值的计算,最终实现精准的机器人飞拍定位。实验表明:较之传统视控分离方案,该机器人飞拍纠偏系统具有简捷易用、精度和效率高等优势。

Aiming at the problem that the traditional robot system for flying-shooting has the separation of vision and control,which leads to the low real-time and poor efficiency of the system,this paper puts forward the scheme of integration of vision and control,and studies the relevant algorithms of flying-shooting.In order to improve the real-time performance of the system,a hard real-time system based on windows is adopted as the operating system.The communication of robot control module and vision module is realized by sharing memory in the same operating system.Based on the improved KALMAN Filtering,a sensorless accurate time trigger algorithm is designed to reduce the positioning error of the robot at the same position when the camera is triggered.The processing speed of image is improved by deep learning,which ensures to complete the calculation of deviation value before the next beat of the robot,and finally the accurate positioning of flying-shooting is realized.The experimental results show that compared with the traditional separate scheme of vision-control system,the robot flying-shooting system for correcting deviation designed in this paper has the advantages of simplicity,precision and efficiency.