Research On Tracking And Grasping Of Moving Objects Based On Visual Feedback Information

In recent years,the research of intelligent robot technology has made great progress,especially in the field of intelligent assembly.But for mobile objects,we still face a series of challenges,such as camera jitter,target occlusion,collision prone obstacles in machine arm motion planning,and early prediction of object trajectories.In order to solve the above problems,this paper carried out an in-depth study,the main results are as follows:(1)Establish the manipulator kinematics model,and calculate the manipulator's workspace,good foundation for the future of mobile platform planning;according to the principle of minimum arm momentum,inverse optimal solution,and according to the actual situation of each joint movement,to determine whether the collision with obstacles,to select the optimal collision free solution.(2)According to the existing problems of the artificial potential field method,such as cannot reach the target point in the process of movement to crash barrier and easy to fall into local minimum,the repulsion function is presented,and the function of gravity increase makes the random jump out the local minimum point method,and the effectiveness of these methods is verified by experiment.(3)A multi feature fusion method is proposed to solve the problem of target occlusion with HOG features.The CN feature is introduced to increase the color feature of the target,which can distinguish between the target and the non target.Through the fusion of two features,the accuracy of target tracking is improved in the case of occlusion.(4)Manual presentation of features often cannot express the essence of the object.In this paper,we use the AlexNet depth network to extract second layers of convolution as the feature of target tracking.A large number of experiments have been made to verify that the depth features have high accuracy and robustness.