Design And Implementation Of Character Detection System For Auto Parts Based On Machine Vision

In the production process of automobile parts,different embossed characters on the surface of different parts are used to identify the product model and production date of the part,which is convenient for the quality control of the part during the production process and the subsequent product traceability.Due to the complexity of the industrial environment,the traditional character information on the surface of parts is manually detected and manually entered into the computer for information management.However,there are many types of parts and large batches,the use of labor will bring about problems such as low efficiency,error-prone,and large labor costs.With the continuous promotion of Industry 4.0,automation and intelligent production have become the pioneers in leading the industry.ERP management systems are introduced into automobile parts manufacturers in large numbers.This paper analyzes the embossed characters on the surface of parts and designs a set of machine vision character detection system of automobile parts meets the requirements of the enterprise’s automated production.The main research contents in the design process are as follows:(1)In the image acquisition stage,the white LED light source and coaxial illumination were selected,combined with the industrial area array camera and lens to complete the part character image collection,to a certain extent,the problem of surface reflection on the part was solved.(2)In the image preprocessing stage,for the problem of low contrast of the part character image,several image enhancement algorithms are studied and compared,and the Retinex algorithm was used to enhance the part character image.Secondly,the bilateral character filtering method was used to filter and denoise the part character images to obtain character images with better quality.Aiming at the inclination problem that is easy to occur in the part feeding process,combined with the surface characteristics of the part,the longest straight line at the upper end of the part was found using the Hough transform method,and the character image was inclined corrected according to the inclination angle of the straight line.(3)In the research of character positioning algorithm,the template matching method based on normalized product correlation was used to complete the positioning of the character area according to the characteristics of the embossed characters of the part and the information of the known characters;The projection method of the single character segmentation can complete the adaptive segmentation of normal characters and sticky characters;In the study of character recognition algorithms,three methods were used to study part character recognition,the recognition results were evaluated in three aspects from recognition rate,recognition speed,and anti-noise ability.Finally,the HOG + SVM method was used to recognize part characters.(4)Build the hardware of the detection system,integrate the above image algorithms,and complete the development of the host computer interface.Select appropriate tilt correction and character positioning methods for various types of parts,and use the system’s character segmentation and recognition methods for part character detection and recognition.The test results show that the system can realize the recognition of multi-part embossed characters and meet the detection requirements of enterprises.