The Pivotal Technology Research On Robot Workpiece Recognition

Industrial Robot visual system percepts the external environment and feeds back the visual information to control system, recognizing and manipulating the target. Effective extraction and matching the Workpiece image characteristics is the key of recognition workpiece goal. However, Workpiece identification and manipulation are easily affected by most complex problems, such as, dust particles, mechanical arm shaking, light, angles and flat degree of the camera film in complex environment. This paper study on the problems of Robot in the workpiece image detection and matching, specific tasks as follows:(1) Machine vision technology and the overview of present situation of robot workpiece recognition:The present study situation and technology of machine vision and workpiece recognition based on machine vision is discussed.(2) The key principle technology analysis of workpiece image feature extraction and merge:The related concept and method principle about workpiece image data extraction and fusion problem is discussed.(3) Based on GA optimization of Histogram of Oriented Gradient and Local Binary Pattern features fusion technology is posed:The basic principle, advantages and disadvantages of HOG and LBP operators is analyzed. In order to avoid defect on the feature extraction in texture image and global workpiece image of LBP and HOG respectively, both the two method extracted image characteristics are concatenated fused. Fusion factors are optimized by GA algorithm to obtain the optimal fusion weights, advantageous to the workpiece recognition.(4) Choosing the model of Workpiece recognition and classification: To verify the workpiece feature fusion result, the introduction of the ELM model and commonly used SVM model and optimize the performance difference between evaluation model of classification, among them, because each classification model parameters are involved, designated by the human to find the optimal parameters, and to ensure that the recognition rate, the introduction of optimization speed, strong global cuckoo algorithm, is used to model and parameter optimization.(5) The test on Workpiece image feature fusion and target identification:First, in the same classification model, combined with multiple denoising methods, verified the fusion feature has a higher rate than independent feature recognition, and then according to the data fusion characteristics, determine the optimal classification model in the model 2 class identified by testing and evaluating, finally, acquired the optimal method and optimal parameter in workpiece image feature extraction and classification identification last, Using the confusion matrix and ROC curve to evaluate model classification performance from different angles.The experimental results indicate that Using GA optimization of LBP-HOG feature fusion technology, and combining the cuckoo optimizing ELM classification model of workpiece based on image recognition rate reach above 95%, and the efficiency is higher, the method proposed in this paper has certain practical and reference value.