机器人视觉Kalman和FIR滤波稳像算法设计与比较

Algorithm Design and Comparison of Kalman and FIR Filter Methods for Image Stabilization of Robot Vision

稳像是提高基于视觉的移动机器人作业精度的关键。论文建立了完整的稳像算法流程,包含图像运动学模型、KLT特征提取、SAD特征匹配和滤波算法;设计了运动参数的Kalman和FIR滤波算法;并利用MATLAB实现了运动参数的Kalman和FIR滤波器;仿真验证和对比分析了Kalman和FIR滤波器对运动参数的去抖效果。结果表明,机器人视觉稳像中,Kalman滤波效果优于FIR滤波。用VC++和OpenCV编程实现了基于Kalman滤波的机器人视觉稳像软件,在双机器人移动平台上开展了实验,稳像计算时间小于视频采样时间,系统满足机器人对接作业实时性和精度要求。

Image stabilization is the key for accurate docking operations of robots with vision. The whole algorithm of image stabilization is established, including images kinematics model, KLT feature pixels detecting, SAD feature pixels matching and filters. Kalman and FIR filters are designed for smoothing images motion parameters and built in MATLAB. Simulation of filter of motion un-intended parameters is implemented to indicate removing jitter effect. Kalman filter is compared with FIR filter. Comparison curves and tables are given, which demonstrate that Kalman filter is better than FIR in robot vision image stabilization process. Based on VC＋＋ and OpenCV, image stabilization software is programmed, and experiments are completed on double moving robots docking operation platform. The algorithm running time is less than the sampling period, and the precision and real-time demands are contented.