Research On Location Method Of Stereo Vision Robot In Dynamic Environment

At present,intelligent equipment such as service robots and AGV are more and more widely used.They have advantages in reducing labor costs,improving user experience.Realizing the above functions is based on the robots odometry algorithm.However,most of the current odometry research is designed based on the assumption of a static environment.In real world,the working environment of robots often contains dynamic objects with absolute speeds such as people and cars.When the robot's perception sensor depends mainly on visual sensor,dynamic objects will seriously affect the robot's own pose estimation and the accuracy of dynamic target detection.High dynamic environment challenges such as a large image proportion of dynamic object and motion of dynamic objects will reduce algorithm estimation Accuracy and robustness.Therefore,the ego-motion of robots and the dynamic feature selection in a dynamic environment is an urgent problem to be solved.The existing vision algorithms cannot solve the problem in high dynamic environment,and the multi-sensor fusion algorithm is too expensive,so this paper proposes an algorithm for solving the robot's pose in a dynamic environment,which is based on the vision method and is supplemented by the low-cost IMU(Inertial Measurement Unit).This paper firstly calibrate the sensor,complete the offline calibration and online calibration of sensor parameters,and then design the loose coupling method of vision and IMU(Inertial Measurement Unit)to identify the dynamic feature points in vision,and use the tight coupling of visual inertia to separate the static feature points and dynamic feature points,bring into the back end to solve the robot pose.Finally,the experiment proves that the algorithm designed in this paper has strong anti-dynamic interference ability and accuracy in both lowdynamic and high-dynamic environments.Specifically,this paper uses the loose coupling of binocular and IMU to obtain the motion vector of the visual feature point,and uses the Mahalanobis distance to consider the uncertainty of the motion vector to realize the recognition of the dynamic feature point in each frame of image.The visual inertial tight coupling algorithm uses the feature points identified in the previous step,and adds the back-end error term of the visual dynamic information as an algorithm improvement.The robot pose is solved using a sliding window to make the robot's positioning in a dynamic environment more accurate.According to the classification and discussion of different factors such as the presence or absence of dynamic objects,the proportion of images of dynamic objects,the motion speed of dynamic objects,and the relative motion of the robot,the positioning estimation experiment of the visual robot under different environments is set up,and the algorithm in this paper is compared with other algorithms on the robot.The experimental results show that the algorithm can effectively identify the dynamic object information in the vision dynamic environment,and can obtain more accurate odometry information of a mobile robot in dynamic environments and static environments,and handles a large number of dynamic targets or high proportion of images in high dynamic situations,the robot has anti-dynamic interference ability.