Research On Multi-sensor Fusion Indoo Localization Of Spherical Robot Based On Image Blur Detection

Spherical robot is a new type of mobile robot with spherical shell and rolling motion.However,when it's rolling,frequent shaking will lead to image blur,which makes the design of localization system based on visual information difficult.In addition,the special mechanical structure makes its wheel speed odometer model different from common mobile robots.The sensor fusion method of spherical robot also has some particularity.Aiming at the above difficulties,this paper designs an image blur detection algorithm according to the characteristics of spherical robot and a multi-sensor fusion indoor localization system for spherical robot based on monocular vision and image blur detection.This paper verifies the effectiveness of the algorithm on the collected spherical robot data set.The main research work and results of this paper include:1.According to the special mechanical structure and motion mode of the spherical robot,the kinematics model is carried out.According to the model,the wheel speed odometry of the spherical robot is realized by combining the wheel speed meter information and inertial sensor data.The inertial sensor and vision sensor of the spherical robot platform are analyzed and the parameters are calibrated.2.Aiming at the challenge that spherical robot images are prone to motion blur,an image blur estimation algorithm based on edge width detection is designed.The adaptive-network-based fuzzy inference system is used to improve the image blur estimation algorithm by adjusting the parameters adaptively and fusing the image-based fuzzy information and inertial sensor data.The image data set of spherical robot platform is established.After experimental verification,the algorithm proposed in this paper has improved some indicators compared with typical methods,and the time-consuming is lower than that of typical methods.3.Through image blur detection,clearer images are selected as key frames,and the multi-sensor fusion indoor localization system for spherical robot is designed and realized.A stepped key frame selection strategy is designed.The key frame selection is performed using the image blur detection result,and the image edge width is used to dynamically adjust the covariance of the visual constraints in the fusion.The experimental results show that the localization system is applicable for spherical robot,which reduces the average image blur of the key frame database,increases the effective loop correction frequency,and reduces the positioning closed-loop error.