Research And Development Of The Real-time Matching Technology Based On Geometirc Primitives

Targeting position technology is an integral part of machine vision. It is an importantprerequisite for the application of workpiece size measurement, recognition and defectdetection of objects, and recognition of characters. Targeting position technology has beenapplied in many fields, such as industry, agriculture, food packaging, pharmaceuticals, andauto production lines. However, there are still many problems in the algorithm like poortargeting accuracy and lack of real-time in the application of industrial production, which isone of the key and difficult problems of today’s research. By studying lots of targeting andmatching algorithms domestic and abroad, this paper proposes a targeting and matchingalgorithm based on geometric primitives according to the actual needs of the work on thepipeline industry in industrial production.For the images captured by smart cameras, this paper studies the extraction method ofgeometric primitives feature. We use an efficient profile tracing rule to tracking the profile ofworkpiece, and use polygonal approximation algorithm and least-squares fitting algorithm todivide the geometric primitives of the profile into linear and circular primitives, and thenobtain the geometric parameters.For the workpieces with arc, this paper proposed a real-time targeting and matchingworkpiece algorithm based on arc primitive. This algorithm produces templates in offlinemode, and extracts the feature of geometric primitives as template information. In onlinesystem, it achieves preliminary locating by using the longest arc primitives as the primitivepriority located, and then calculates the relative distance between primitives to finish thematching of the remaining primitive to finish the targeting.To solve the problem of workpiece targeting in assembly line, this paper proposes ageometric hashing method targeting and matching algorithm based on straight line primitives.This algorithm extracts character of straight line primitives, and use coordinate to expressstraight line primitives in off-line mode, and establishes geometric hash table through therelationship between linear primitive. In online system, it completes the targeting according toinformation query hash table of linear primitive of the actual images.According to the practical need of industrial production, we collect various types testimages of different work environments to verify the accuracy and real-time performance ofthe algorithm proposed in this paper. The experiments show that, the extracted feature ofgeometric primitives basically meets the demands of production, and the running times of thetwo targeting algorithms are the millisecond with more than95%recognition rate. Therefore,the algorithm proposed in this paper could equal to the demand of real-time and accuracy inindustry field.