Research On Character Recognition Method Of Disc Casting Based On Photometric Stereo Technology

As the advance of Industry 4.0 in China,the development and maturity of automobile industry makes some related industries grow vigorously.Promoting production management and controlling the products quality,it becomes an essential method which casts batch number characters on products and traces the source at the manufacturing site.This thesis aims at the batch number of production line for disc casting,which identifies the problems includes the long system response time,high recognition error rate and easy confusion.Based on the photometric stereo technology,it proposes a method of disc casting character recognition,which can effectively avoid the characters confusion which was caused by the low contrast between casting characters and background and unrecognized or misrecognized batch numbers caused by disorderly placement of products during processing.In order to the intellectualization of the manufacturing process,the main content as follows:First,according to the on-site technical demands,this thesis designs the completed project of characters identification system for disc casting.On the basis of the technical index,the selection of devices and equipment in the character identification system of disc casting is determined,which builds a character identification system for disc castings based on machine vision.Second,directed at the problem that the same color of the character zone and background of disc casting and low contrast.Combined with the photometric stereo technology,it uses a gradient data processing method based on curvature calculation to reconstruct the image and utilizes Gaussian filter to denoise the reconstructed image.Based on prior knowledge,it proposes a specific character positioning method for disc castings.Radon transformation method is rectified and oriented to the character zone,split the single character by projection analysis,and finally the character sample is normalized by linear algorithm.Third,in the process of character recognition of disc casting,the current recognition method is ineffective due to the confusion of character rotation and similar structures.Combining partial HOG features and rotation invariant LBP feature extraction method,it proposes a fusion feature method of partial HOG and multi-layer rotation invariant LBP.Due to the few character samples of disc castings,it selects support vector machine as a classifier.By the experiment,the fusion feature was mentioned in the thesis can effectively solve the problem that is character rotation and similar structures.Last,this thesis designs a character recognition system for disc casting,achieves the functions of light-source control,image capture,disposal and analysis as well as human-computer interaction.Verifying the character recognition system on the software platform,the results show that the gradient data processing method used is excellent.It proposes a local HOG and multi-layer rotation invariant LBP fusion feature,and combined with a support vector machine classifier,which can effectively reduce the rate of error identification of confused and rotated characters and fulfill the project index requirements.Thus,the disc casting character recognition system based on photometric stereo technology designed in this thesis achieves the desired results,it can be applied in the process of brake disc processing to improve product quality in the manufacturing industry.