Jitter compensation using fuzzy rules and Kalman filter for mobile robots vision

Stable vision is the key for docking operation of mobile robots. Based on affine transformation image kinematics and recurrence relations, the sub-region KLT feature extract was designed, and fuzzy rules was established to get potential feature windows. Optimization of sum of absolute difference was used to match feature points. Diamond search template was adapted to make matching fast, and adaptive varying template size scheme was proposed to solve the problem the minimum SAD is not exclusive. Through solving over-determined image kinematics equations using the least squares algorithm, got the motion parameters. According to the observation model of intended motion, unintended motion was eliminated by Kalman filter. Then using filtered parameters to reconstruct image and designing bilinear interpolation algorithm to deal with non-integer pixel coordinate produced by flaot computing. On the autonomous robot experiment was completed. Results method proposed meet requirements of accuracy and real time.