Research On Wheeled Robot Global Localization And Target Tracking With Recognition

In recent years,wheeled intelligent robots have attracted wide attention from the scientific community in many countries.Many research institutes and enterprises have carried out many technical competitions for wheeled robots,among which the ICRA Artificial Intelligence Challenge is one of them.This paper focuses on the technical requirements of the artificial intelligence challenge,combined with the intelligent and practical prospects of wheeled robots,in the indoor global localization of wheeled robots,the recognition,tracking and distance resolution of visual target features,and the reduction of control system disturbances.the study.(1)Global localization of wheeled robots is the basic condition for robotic autonomous movement.In this paper,the trajectory estimation algorithm is adopted,and the cumulative moving pose of the robot is captured by the orthogonal wheel.The omnidirectional wheel moving model is used instead of the differential wheel moving model in the process of establishing the pose updating model,combined with the radar environment measurement.Data update the current pose of the robot.The experimental results show that the average deviation of the improved positioning point is 0.0494 m,and the standard deviation between the measured value and the true value is 0.026 m.The improved global localization model and pose acquisition method can improve the positioning accuracy of the robot.(2)For the special recognition target of armor plate,a recognition scheme combining color features and digital features is established.Color channel separation and morphological processing methods are used to extract color channel features and identify the recognition area.Using the direction gradient feature,the support vector machine and the correlation filter algorithm are used for digital recognition and target following.The Zhang Zhengyou calibration method is used to determine the internal parameter matrix of the robot,and the pnp solution method is used to calculate the pose of the target armor plate relative to the camera.Through experiments,the correct rate of identification is 98.94%,the maximum relative error of the distance solution is 0.9%,and the maximum relative error peak is 4.61%,which meets the actual requirements.(3)For the two-axis gimbal control scheme,the Active Disturbance Rejection Control method is used to improve the proportional-integral derivative control scheme.The dynamic model of the two-axis gimbal is established,and the dynamic model is transformed into the control differential equation.The linear differential tracker is designed and extended state observer adjusts the feedback control rate related parameters to meet the criteria for rapid and accurate response.The experimental results show that the experimental data shows that the follow-up response time is less than 0.1 s and the amplitude following error is less than 1°.It is obviously better than PID control for static disturbance auto-disturbance control,and can be better than PID control for dynamic disturbance auto-disturbance control.(4)The experiments of indoor full field positioning,target feature recognition of visual armor plate,tracking and ranging,and Active Disturbance Rejection Control method used for gimbal are carried out on the real wheeled robot.The validity of the algorithm is verified.