Research On Target Location And Tracking Method Of Wheel-Propeller Robot

In recent years,the state has paid more and more attention to the oceans,which occupy 70%of the earth,and various institutions have been increasing the development and exploration of the oceans.The research on the coastal environment mainly includes biological observation,environmental detection,rescue and rescue,hydrological observation and so on.With the increasing demand and scope of operation in this area,the traditional stand-alone operation capacity is limited and inefficient,and formation cooperation is more efficient and forward-looking.The ability of robots to perceive each other’s motion state in real time is the basis and key content of their cooperative work.In this paper,the designated model is used as a substitute for the amphibious paddle robot,and the research work on the target location and tracking method of the robot is carried out to improve the autonomous operation ability of the amphibious robot in order to meet the increasing demand of operation in the coastal environment.The basic theory of target location and tracking method of wheel propeller robot is studied.In this paper,based on the research platform of amphibious wheel-propeller robot,the kinematics model of wheel-propeller robot is established,and the measurement coordinate set and transformation relation matrix of optical vision sensor of amphibious wheel-propeller robot are established.After the objective evaluation of the commonly used underwater image enhancement methods,the enhancement results and the CLAHE enhancement method with better stability are finally selected.Facing the actual task requirements,the target location and tracking methods are studied from three aspects:target recognition method,target tracking method and target global motion state estimation method.Aiming at the problem of target recognition of wheel-propeller robot,the methods of positioning and template acquisition of designated targets are studied.Based on the target recognition principle of the YOLOv4-Tiny network model,this paper adds the channel attention mechanism to the network model,and strengthens the useful feature channel and weakens the useless channel according to the value of its weight.The test results on airborne embedded processor(TX2)show that this method improves the recognition accuracy while ensuring that the forward propagation speed of the model meets the real-time requirements.Aiming at the problem of target tracking of wheel-propeller robot,the method of target location and tracking in continuous frames is studied.Based on the target tracking principle of the twin network model,this paper optimizes the network model by using the depth separable convolution formed by the linear combination of point-by-point convolution and depth-bydepth convolution,which effectively reduces the number of parameters when the real-time input image propagates forward in the candidate network branch.The test results on airborne embedded processor(TX2)show that this method effectively increases the real-time performance of the tracker,and improves the average tracking accuracy and tracking success rate.Aiming at the problem of target tracking of wheel-propeller robot,the method of target location and tracking in continuous frames is studied.Based on the target tracking principle of the twin network model,this paper optimizes the network model by using the depth separable convolution formed by the linear combination of point-by-point convolution and depth-bydepth convolution,which effectively reduces the number of parameters when the real-time input image propagates forward in the candidate network branch.The test results on airborne embedded processor(TX2)show that this method effectively increases the real-time performance of the tracker.The software and hardware of the control system of the wheel-propeller robot are built.the system is divided into a hierarchical progressive hardware control system and a parallel PD target tracking control strategy.The system can process multi-channel data information in real-time and parallel and guide the wheel-propeller robot to track moving objects.An experimental study on the maneuverability of an amphibious wheeled propeller robot was carried out in a simulated coastal(land and underwater)environment.The experiment of locating and tracking the moving target of the wheel-propeller robot is completed,and the effectiveness of the target location and tracking method based on the wheel-propeller robot proposed in this paper is verified.