Research On Random Distribution Of Workpiece Automatic Sequencing Based On Machine Vision

Intelligent manufacturing is the future strategic development direction of the country.Industrial robots are important equipment for intelligent manufacturing,and they are used more and more widely in all walks of life.Object recognition is a prerequisite for industrial robots to replace manual work.Accurate recognition and grasping of industrial robots has become one of the difficult problems in automated manufacturing among objects with different shapes and sizes.Machine vision,as an assistant technology of industrial robots,is widely used in various fields,such as industry,agriculture,medical and health,aerospace and so on.Visual perception technology has become a research hotspot of scholars at home and abroad.The combination of machine vision and laser sensing to perceive the feature information of target object can solve the problem of target recognition and location quickly and effectively.It is of great significance for industrial robots on automatic production lines to automatically identify and locate target objects and implement automatic grasping.This paper chooses the workpieces with automatic sequencing requirement on the automatic production line as the research object.Some current mainstream target detection methods and traditional image segmentation algorithms are studied.After comparing and analyzing the effect of the algorithm,the optimized algorithm and laser sensor information fusion are selected to study the fast target recognition and location algorithm.The main contents of this paper are as follows:1)A workpiece's image acquisition platform was built and the workpiece's image information was collected.Two kinds of traditional workpiece's localization algorithms are studied: threshold segmentation based localization algorithm and Ada Boost based localization algorithm.In the location algorithm of threshold segmentation,the original color image is transformed into YCb Cr mode based on the color feature of the workpieces.The contrast of workpiece's image is enhanced by image preprocessing.Ostu algorithm is used to binarize the image and further morphological processing of the extracted region is carried out to obtain a smoother and complete workpiece region.In Ada Boost localization algorithm,local binary patterns(LBP)features are selected to train cascade classifiers and compared with Haar features and HOG features.It is found that LBP feature training cascade classifier is effective.Finally,the effect of two traditional workpiece's location algorithms is compared.It is found that the two kinds of workpiece's localization algorithms have good detection effect in the case of simple placing,but in the case of complex background and stacking of workpieces,the detection effect of the localization algorithm based on threshold segmentation is poor,and the workpiece's localization can not be achieved.Although the localization algorithm based on Ada Boost can segment the workpieces,there are some problems such as incomplete segmentation,wrong segmentation and large positioning deviation.2)The convolution neural network algorithm in deep learning is studied and several main target detection algorithms in recent years are analyzed.The SSD convolution neural network algorithm is proposed to locate the workpieces.The pretreatment of workpiece's samples and label making are introduced,and training samples are input for SSD model training.The detector is acquired and the image acquired by vision is detected by workpiece's location.The SSD algorithm is optimized by modifying the size of the input image and data amplification technology.After constant adjustment of the algorithm,the correct recognition rate of the workpieces reaches 96.4%.3)A rotation template matching algorithm based on gray level is proposed to detect the horizontal rotation angle of workpiece.The ROI region obtained by workpiece localization algorithm is filtered by image preprocessing,morphological processing,and connected region screening.Then,the rotation template matching algorithm based on gray level is used to detect the horizontal rotation angle of the workpieces.This method can effectively acquire the image of the workpieces,ensure the effect of template matching,and make the rotation angle detection of the workpiece more accurate..4)A set of active tracking laser sensing system is built.Through servo drive,the normal vector of workpiece's position and attitude can be obtained by combining multiple laser sensors.Comparing with single laser detection,the system can obtain the normal vector of space position and attitude at one time.This method retains the accuracy of laser detection,and effectively reduces the detection time and improves the efficiency.5)Grabbing experiments are carried out to verify the effectiveness of the algorithm.The algorithm achieves good results in different scenarios.The recognition rate is 96.1%,the grabbing rate is 96%,and the detection time of a single workpiece is within 0.5 seconds.This provides a feasible means for automatic grasping of randomly distributed workpieces on automatic production lines.