Positioning And Grasping Technology Of Small Parts Of Automobile Based On Visual Guidance

With the rapid development of machine vision technology and the continuous expansion of application field,especially in the industry,it has become one of the tools to obtain information and a key technology reflecting the intelligence of industry 4.0.Based on machine vision,this paper studies the positioning and grasping of complex and diverse automobile small parts in an unstructured environment.In terms of content,it includes the recognition,classification and positioning of the target,as well as the path planning design of the robot in the process of grasping.From the difficulty point of view,there are two difficulties for the robot in the target recognition: first,when many small parts of the car are placed in the basket intricately,there is a large amount of unknown information due to the complexity and diversity of the types and mutual shielding.Second,obtain the threedimensional information of the target group,and determine the grabbing order and pose of the target.In addition,the traditional robot operation is to locate and grab the known existing targets through the repeated teaching calibration of the teaching demonstrator,and the execution mode is relatively rigid,and the generalization ability is not enough,so it is difficult to meet the requirements of industrial robots for multi-type and multi-target operation.Therefore,this paper proposes a method that is suitable for positioning and grasping of small auto parts,which is studied in the following three aspects:First identification of a small car parts for complicated unknown environment classification problem,this paper proposes an adaptive rate convolution model of neural network training,the neural network combined with Faster R-CNN,the target area and judgment that the IoU candidate area are proposed,at the same time increase the constraints on judgement criterion is presented in this paper,fuzzy regression classification for correction.Experiments show that the proposed method can correct the errors properly and reduce the error rate when the objects are shielded from each other.Secondly,in view of the problem of stereo information data acquisition,this paper analyzes the principle and shortcomings of binocular parallax method,and puts forward local parallax and multi-line scanning parallax method.Through experimental comparative analysis,this method can complete the parallax map acquisition in real time,and has certain improvement in time and accuracy.Finally,aiming at the design problem of robot path planning in the grasping process,this paper first analyzes the advantages and disadvantages of parabola method and polynomial method,and then proposes a combined interpolation method to plan the path and to grasp the captured workpiece.By building a visual robot integrated system platform and conducting simulation experiments in a complex environment,the cost is low and the effect is good.Through the final analysis,the research work of this paper has great reference value and application value for the positioning and grasping of small and medium-sized parts.