Research On Key Issues Of Styling Automation For Automobile Frontal Faces

With the increasing competition in the passenger car market,while ensuring the functional quality of the car,the family characteristics of the brand and novelty of the car’s styling have gradually become the core factor of the car’s competitiveness.As one of the core differentiated design areas of the automobile,the front face shape of the automobile has received extensive attention in enhancing brand competitiveness.Realize the automation of automobile front face modeling,while maintaining the family design,it can provide more diversified automobile front face modeling,thereby improving the efficiency of automobile concept design.Due to the high complexity of the topological complexity of the front face of the automobile and the uncontrollability of design semantic editing,it is very difficult to automate the design of the front face of the automobile.This paper conducts a series of big data-driven research on the key technologies of automobile front face modeling automation,through the sorting and classification of automobile front faces,creating a large-scale automobile front face modeling image library,automatic front face type recognition and key area detection,and key modeling features The automatic point extraction and the deformation method driven by the corresponding feature points realize the design space roaming of the front face of the car.The specific content is as follows:Topological analysis of the front face of the car,automatic recognition and area detection based on deep learning network.Aiming at the topological complexity of automobile front face modeling,this paper,based on Gestalt psychology theory,organizes and analyzes a large number of automobile front face image data,divides the automobile front face modeling into five categories,and creates the corresponding automobile front face based on the classification results Classification and detection data set AFCD4000.On this basis,the YOLO V5 l algorithm is used to automatically recognize and detect the front face of the car.Numerical experiments show that the classification recognition and detection method used in this article has an average accuracy of 99% with an intersection ratio of 0.5 to 0.95,which is effective and robust.Automatic extraction of key feature points on the front face of a car based on deep learning.On the one hand,the key feature points of the front face of a car can reflect the essential styling characteristics of the car model,and on the other hand,it can be used as a handle element for image deformation and deformation.Through collation and analysis,this paper defines 44 unified key feature points of the front face of the car,covering the key elements of the core shape of the front face of the car,and creates a key feature point data set of the front face image of the car on the basis of AFCD4000.Furthermore,this paper proposes a cascaded DANRes Net50 feature point extraction algorithm to automatically extract key feature points of the front face of the car.Numerical experiments show that the average error of the feature point extraction method in this paper is 0.0149,which is 6.2% lower than the original.Design space roaming of car front face based on image deformation.For the input car front face image sequence,first perform the car front face classification and target area detection,and then obtain the key feature points of the car front face through the automatic extraction method,and use the key feature points as the driving elements to generate the deformation control grid,which is driven by the grid The local affine transformation of the image realizes the roaming of the design space of the front face of the car,and then obtains a rich and diversified front face shape of the car while maintaining the family characteristics of the car.In addition,this paper also conducts a series of exploration researches on the cross-category,crossbrand,etc.of the car front face design space to further expand the car front face design space.