Research On Human Object Visual Following Method Of Walking Robot

With the development of society and the progress of technology,demand for robots increases rapidly in daily life and manufacturing.Following robot is one of the most important development directions.Transportation tool with the technology of following robot can greatly improve safety and convenience when going out,especially for the old and disabled.Therefore,a walking robot has been designed in this paper.And the longterm target following technology in the outdoor interference scene has been studied based on the low configuration robot.In this paper,a service robot platform with two functional modes called walking is built to meet the travel needs of special groups as the carrier of subsequent experiments.In order to adapt to the different features of each functional mode,the variable drive chassis and body folding mechanism of the walking robot are designed and built specifically,and the stability of the folding body is checked.On this basis,the control system of the walking following robot is designed: Firstly,the hardware system framework is determined and the main hardware modules are selected;and secondly target tracking and motion control modules ard developed based on ROS system,and the construction of robot software system is completed.A more robust long-time target tracking algorithm is proposed for the actual use of the walking robot.The limitations of the kernel correlation filtering algorithm in outdoor application are analyzed and improved from two aspects of input sample training and frame fusion.The threshold of image light quality judgment is proposed and the light intensity distribution is corrected.The complementary characteristics are characterized by adaptive fusion based on the characteristic entropy,and the sample quality is improved by using the binary mask matrix to improve the training effect.In addition,based on the multi-threshold confidence judgment mechanism,the model update and target rerecognition modules are added to the tracking framework to deal with the error accumulation and target loss in the algorithm's long-term tracking process.Ablation experiments are carried out on the OTB100 data set to verify the effectiveness of each module,and then the comprehensive improved algorithm is compared with other classical correlation filtering algorithms.Qualitative and quantitative analysis experiments proved that the tracking effect of the improved algorithm is significantly improved.In order to improve the stability of the robot in the process of moving,this paper proposes a following control strategy based on fuzzy control.A monocular vision observation model is established,and the target position acquired by the vision tracking algorithm is converted into the relative pose in the robot coordinate system.The following kinematics model of human target is established,the velocity controller and its following control law are designed according to the inversion idea,and the fuzzy controller is designed to smooth the tracking stability of the robot.The experimental results in simulation following environment and robot entity test show that the following controller designed in this paper meets the design requirements.