Research On Following Technology Of Two-Wheeled Manned Self-Balancing Vehicle

With the continuous development of robot technology,following robot is becoming one of the research hotspots.The following robot means that the robot does not have any physical contact with the following target in the process of autonomous following and keeps a certain distance between them.The two-wheel self-balancing vehicle can belong to the field of intelligent robot,which has the characteristics of small occupancy and high flexibility.The research on the following technology based on the two-wheel self-balancing vehicle will have more flexible application than the traditional four-wheel robot or tracked robot.Based on this opportunity,this paper designs a two-wheel manned self-balancing following vehicle.The main research contents are as follows:(1)Based on the hardware of the two-wheel manned self-balancing vehicle of our laboratory and the monocular vision following scheme,a manned following balancing vehicle system was constructed.(2)An improved adaptive Kalman filter algorithm is designed for attitude fusion of balanced vehicle.Conventional Kalman filtering algorithm sets system noise and measurement noise as constant values,which may cause filtering divergence due to the variation of measurement noise in practical application.In this paper,Sage-Husa adaptive Kalman filter algorithm is improved,which ensures filtering accuracy by real-time estimation of measurement noise variance,and the divergence problem of filtering is solved by combining the strong tracking Kalman filter.(3)To solve the problem of target detection,the April Tag algorithm suitable for embedded platform is adopted,and the April Tag target tag is taken as the followed target.On this basis,a secondary target detection algorithm is designed to deal with the local occlusion of the following target,and the recognition effect is improved.The correct recognition rate is 98.4% when the target tag is not occluded,and 91.2% when the target tag is partially occluded.Aiming at the situation that the target may be lost,the prediction effect of the least square curve fitting algorithm and the Kalman filter algorithm on the target position is compared,and the Kalman filter algorithm is finally selected to follow the target position prediction.(4)Aiming at the problem of motion control,a fuzzy RBF neural network PID control method is proposed.In view of the defects of the current balanced vehicle control system,the control parameters are fixed and can not be self-tuning,a fuzzy neural network controller is designed to dynamically adjust the parameters.The traditional PID control,fuzzy PID control and fuzzy RBF neural network PID control methods were used to carry out comparative tests.The results show that the system overshoot reaches 31% and the vibration is severe when the traditional PID control is used.When the fuzzy PID control is adopted,the overshoot of the system is 14%,and the vibration is obviously weakened.When using fuzzy neural network PID control,the overshoot is close to zero,the system has no shock,and the self-adaptability and anti-interference are stronger.(5)The test experiment was carried out based on the hardware platform of the manned self-balancing vehicle following system.The experimental results show that the target recognition algorithm designed in this paper is not affected by illumination intensity,and can identify the target quickly and accurately under the condition of target tag tilt and partial occlusion.It has good robustness and can be applied to complex following scenes.The system can achieve a stable tracking effect to the target tag at a uniform speed,and the tracking system has a certain anti-interference ability.