Bin Picking Based On Monocular Stereo Vision

Bin-picking is a typical application of combination of machine vision and robot,and also a hot issue which is wildly researched.The main binpicking solutions include monocular vision,binocular vision,structured light,laser ranging and so on.Compared with other schemes,monocular vision is very sensitive to the contour changes caused by occlusion,the reflection of light and shadow.There is less the amount of information available in monocular vision.It's difficult to obtain the complete distribution of top workpieces.In this paper,a kind of light source layout scheme,which can obtain the full visible edges,has been designed and a characteristic surface recognition,which reduces the influence of occlusion on the integrity of the workpiece.Firstly,the distribution characteristic of position accuracy of monocular stereo vision system in the field of view is analyzed.It has been discovered that the resolution of Z direction is close to zero in the view center and increases with the increasing of distance between test point and view center.Secondly,a method is explored to obtain the full visible edges in images.It has been compared and analyzed in this paper the effects of different light arrangements on the workpiece stack image to extract more visible edges.This paper obtains complete visible edge images can be obtained by combining multiple edge images which have highlighted edges of the top pile workpieces into one edge image.Thirdly,the problem that it is difficult to ensure the integrity and accuracy of the workpiece identification has been solved.This paper proposes a new method which is called characteristic surface recognition to obtain more top workpieces.However it also results in the repeated identifications and error identifications of workpiece.In order to provide priority reference to eliminate the repeated identifications and error identifications of workpiece.This paper presents a new method to identify the authenticity of workpiece based on the spatial aggregation degree of the recognitions and the representation of feature surface.Based on the experimental and theoretical analysis,a method for evaluating the degree of spatial aggregation and feature representation of the workpiece is proposed.Fourthly,the field of view is rebuilt according to the visual processing information.The real top workpieces is obtained after eliminating repeated identifications and error identifications of workpiece based on authenticity sorting of identifications.The identification of workpiece,known and unknown areas of background is reconstructed in UG simulation environment.The final picked workpiece is determined after analysis of interference pick-up process based on the results of three-dimensional reconstruction.This paper designs experiments about 50 sets of workpiece piles.The experimental results show that the visual system can get most of the top workpieces and the average visual processing cycle is 27 s.Therefore,the model-based monocular vision scheme proposed in this paper can be applied to Bin-picking system.