Research On Application Of Filter Based Visual Inertial Odometer In Foot Robot

With its superior mobility,adaptability to the environment and multi-function in complex environments,foot robots have become intelligent creation machines for human beings to challenge nature and explore unknown fields.Quadruped robots,as an important branch of foot robots,have been gradually applied in many fields,such as military reconnaissance,blasting,rescue patrol,commercial transportation,public entertainment and so on,and have received widespread attention of scholars.However,when the robot is working in the field,there are still some problems such as inaccurate positioning,obstacle avoidance,and unable to achieve the optimal choice of real-time road environment,which often lead to the lost of the robot,trapped machine,explosive machine and so on.Therefore,the Optimization Research on autonomous positioning and intelligent navigation of robots is the key and difficult point to realize remote intelligent control of robots.At present,Visual Inertial Odometer(VIO)technology has prominent advantages in solving the above problems:its core is to integrate the inertia information of the robot itself and the visual information collected by random camera,and to effectively integrate the two,which not only improves the flexibility and stability of the robot itself,but also ensures the robot's rapid movement or lack of features.Accurate positioning under conditions.In this paper,a visual inertial odometer based on Iterative Extended Kalman Filter is proposed for quadruped robots.Specific research contents are as follows:Firstly,a quadruped robot model is constructed.The motion of the robot is analyzed by D-H parameter method.The Jacobian matrix of the single leg motion of the robot is obtained,and the corresponding relationship between the foot trajectory and the joint angle is deduced.The trot gait motion of the whole robot is simulated and analyzed,and the motion law of the robot is obtained.Secondly,several common feature extraction algorithms are compared,and FAST(Features From Accelerated Segment Test)feature extraction algorithm is selected to extract the key points in the image.Then,image blocks are acquired and filtered according to the extracted feature points,which are used as descriptors of landmarks.Then multi-level image pyramids are extracted from the selected image blocks,and the landmarks are correlated with multi-level image blocks.The photometric errors obtained after image processing are integrated into the iterative update step of the extended iterated Kalman filter,which makes the system more compact.The state vector of the system is defined and the coordinate system centered on the robot is used to reduce the non-linear error.Thirdly,through the modeling of the inertial sensor,the kinematics formula is constructed,the error propagation equation of the state vector is deduced,and the error equation is used as the prediction update of the covariance matrix.At the same time,the camera pose output by the visual odometer is used as the observation to measure and update.This paper presents a method to solve the problem of inconsistent sampling frequency by directly adding the data of inertial measurement unit.This method has less computation,good real-time performance,and does not cause performance loss to the system.Finally,experiments are carried out on open data sets,indoor and outdoor robots and foot robots respectively,and the experimental results are analyzed.Experiments show that the visual inertial odometer based on filter frame can achieve high precision and robustness positioning,and meet the real-time and robustness requirements of the foot robot.