The Research Of Method In Conveyor Tracking Based On Visual Object Tracking And Real Time Guidance

The paper focuses on the appearance quality online inspection of product on roller conveyor.The key problem is the robot tracking: without stopping the conveyor,the robot tracks the moving target with the specified position and attitude and finishes the quality inspection.A conveyor tracking method based on visual objects tracking and real-time guidance is proposed in this paper.The main research contents of this paper are as follows.The conveyor tracking system is built and is calibrated,consisting of a roller conveyor,a 6DOF robot and an external fixed camera.The camera’s view coincides with the working range of the robot.The moving objects on the conveyor are tracked by the fixed camera in real time.Then,the robot is guided to track those objects.A lightweight and efficient multi-object tracking(MOT)algorithm is proposed.Firstly,a object detection algorithm is used to detect all those objects in the current view of camera.Secondly,predicted objects are obtained from the tracking objects in past video.Thirdly,the detected objects in current and the predicted objects from past are associated.At last,the tracking objects are updated with the detected target which is associated with it.Running on CPU,the MOT algorithm efficiency can be better than 50fps@640×480.A control method that process the robot online is proposed.Firstly,the method chooses one of the tracking objects from MOT to be the inspected object.Secondly,the 6D pose of the inspected object is estimated in real time,and the robot working point is also solved.At last,the robot is processed to move smoothly from the starting point to the working point and keeps relatively stationary with the inspected object.Based on the condition that the end-effector of robot should keeps coincident with the working point on object during tracking,the paper drives four evaluating indicator for tracking method,including the position and posture bias between end-effector and working point,position and posture variance of end-effector.The experiments is derived to evaluate the performance of our tracking method.It shows that: 1)in the condition of object’s speed is less than 1m / s,the position bias between end-effector and the working point is less than 80mm;the position variance of end-effector is less than 5mm;the posture bias between end-effector and the working point is less than 4°;the posture variance is less than 1°;2)compared with the existing method,our method does not accumulate the bias during tracking,which is more suitable for the continuously tracking.