Design And Implementation Of Domain-specific Language For Computer Vision

Recently,the technical breakthroughs in various fields have prompted the rapid development of Computer Vision(CV).A lot of efficient operators have been designed and computer vision based softwares and systems have been widely applied.During the development of CV related applications,communication between domain experts and developers has always been the major problem.At the same time,it is difficult to com-bine and call various CV operators.As the main programming language for computer vision,the complexity of C++has greatly reduced the work efficiency of developers.To tackle the problem,we design and implement Computer Vision Domain Specific Language(CVDSL),based on the Xtext framework,adopting a model-driven devel-oping principle.CVDSL,which has simple syntax rules,integrates commonly used CV operators and provides a unified platform for developing applications in computer vision.We first analyze the domain features and extract a meta-model for computer vision with Eclipse Modeling Framework(EMF).Then we define syntax rules for CVDSL with the Xtext framework by the metamodel mapping and rule adding.In order to improve the developing efficiency,we implement a syntax checker which enforce the predefined rules,thus avoiding potential syntax errors.Finally,we implement a code generator for CVDSL which can convert it into C code,achieving the conversion from a domain-specific language to a general programming language(GPL).Through experiments,we compare C++and CVDSL based on two indicators:ba-sic(functionality,ease of use,efficiency)and generated C++code quality.We show that compared to C++,CVDSL is easier to learn.In the development of CV application model,using CVDSL is more efficient and requires less code.The C++code generated by CVDSL code generator is also more standard and concise.Through targeted exper-iments,we proved that CVDSL has good scalability.We also evaluate the CVDSL in detail with the Framework for Qualitative Assessment of DSLs(FQAD),in functional applicability,availability,reliability,scalability,integration,maintainability,produc-tivity,compatibility,expressibility and reusability.We compare CVDSL with Vision-Pro and Halide based on the above indicators and find CVDSL has its own advantages in certain aspects.